A comprehensive, curated survey of Non-Line-of-Sight Imaging research
Authors: Ruixu Geng · Yang Hu · Yan Chen
| Link | |
|---|---|
| Original Survey (peer-reviewed, APSIPA TSIP 2022) |   |
| Updated Survey (2022–2026 extension, not peer-reviewed) |  |
| Interactive Project Homepage |  |
Non-Line-of-Sight (NLOS) imaging reconstructs or senses hidden scenes that cannot be directly observed — for example, around corners, behind walls, inside occluded spaces, or through scattering/opaque media. The key idea is that light, sound, radio-frequency waves, or other signals interact with visible relay surfaces, hidden objects, tags, or environmental reflectors; the indirect measurements encode recoverable information about hidden geometry, albedo, motion, identity, or position.
Emitter / sensor ──► relay wall / reflector / aperture ──► hidden target
▲ │
└──────────── indirect return / transient / RF / sound ◄─┘
measurement → physical forward model → inverse solver / neural prior → hidden scene
Why it matters: autonomous driving, robotics, search-and-rescue, medical imaging, surveillance, and scene understanding all benefit from sensing beyond direct line of sight.
- Latest Additions
- Milestone Timeline
- Taxonomy
- Active NLOS Imaging
- Passive NLOS Imaging
- Deep Learning for NLOS
- New NLOS Scenes and Modalities
- Datasets and Open-Source Code
- Related Surveys and Benchmarks
- Citation
- Contributing
Update run: 27 August 2026. This section tracks newly found or newly completed entries that were not explicitly covered in the previous README / homepage snapshot.
| Year | Paper | Venue / Status | Why it matters |
|---|---|---|---|
| 2025 | Influence of some acquisition parameters in non-line-of-sight imaging — Christnacher et al. | Proc. SPIE 13678, Emerging Technologies and Materials for Security and Defence 2025, 136780E | Systematically studies practical SPAD/transient acquisition choices—including pixel calibration, exposure time, frame count, camera FOV, and focus—to map the speed–transient-quality trade-off before f-k, back-projection, or phasor-field reconstruction, complementing algorithm-side acceleration with acquisition-side optimization. |
| 2025 | Combined geometric and physical optics analysis of passive non-line-of-sight light-field measurement — Sasaki, Grossman, Leger | Optics Express 33(19), 39194–39217 (2025) | Unifies geometric-optics BRDF and physical-optics Wigner analyses to predict how roughness, wavelength, diffraction, and lens focusing govern retrievable passive NLOS light-field information in LWIR and THz regimes. |
| 2022 | Estimation of the 3D spatial location of non-line-of-sight objects using passive THz plenoptic measurements — Sasaki, Grossman, Leger | Optics Express 30(23), 41911–41921 (2022) | Measures spatial and angular THz radiation reflected from rough walls with a room-temperature sensor and refocuses the resulting plenoptic data to recover the 3D location of hidden human-like thermal targets. |
| 2022 | Passive Terahertz Non-Line-of-Sight Imaging — Grossman, Sasaki, Leger | IEEE Transactions on Terahertz Science and Technology 12(5), 489–498 (2022) | Demonstrates passive 336-GHz NLOS human imaging from ordinary rough-surface reflections using uncooled direct-detection hardware, recovering target location, orientation, and pose even for weak specular components. |
| 2021 | Passive 3D location estimation of non-line-of-sight objects from a scattered thermal infrared light field — Sasaki, Hashemi, Leger | Optics Express 29(26), 43642–43661 (2021) | Scans an infrared camera to form a scattered LWIR light-field cube, separates the weak information-bearing component, and localizes human-temperature hidden objects in 3D in a life-size diffusive hallway. |
| 2021 | Non-line-of-sight object location estimation from scattered light using plenoptic data — Sasaki, Leger | Journal of the Optical Society of America A 38(2), 211–228 (2021) | Derives depth/transverse localization limits from scattering physics and noise, then uses a projection-slice/light-field formulation and mixed-space-frequency filtering to estimate hidden-object location near the theoretical resolution limit. |
| 2020 | Light field reconstruction from scattered light using plenoptic data — Sasaki, Leger | Journal of the Optical Society of America A 37(4), 653–670 (2020) | Provides the scattered-light plenoptic inverse foundation: a BRDF-aware Fredholm light-field model, regularized reconstruction, and limits on retrievable hidden-scene information after wall scattering. |
| 2019 | Corner Occluder Computational Periscopy: Estimating a Hidden Scene from a Single Photograph — Seidel et al. | IEEE ICCP 2019 | Uses a known wall edge as a natural aperture so one ordinary-camera penumbra photograph can jointly estimate floor albedo and a 1-D angular representation of the hidden scene. |
| 2020 | Multi-Depth Computational Periscopy with an Ordinary Camera — Saunders et al. | IEEE ICASSP 2020, 9299–9305 | Extends ordinary-camera computational periscopy to a multi-depth hidden scene, recovering two hidden images and their wall-relative depths from a single photograph. |
| 2021 | Two-Dimensional Non-Line-of-Sight Scene Estimation From a Single Edge Occluder — Seidel et al. | IEEE Transactions on Computational Imaging 7, 58–72 (2021) | Adds range to edge-occluder angular sensing, reconstructing a 2-D hidden scene from one penumbra photograph with radial-falloff modeling, inversion algorithms, and Cramér–Rao analysis. |
| 2021 | Fast Computational Periscopy in Challenging Ambient Light Conditions through Optimized Preconditioning — Saunders and Goyal | IEEE ICCP 2021, 1–9 | Uses optimized preconditioning to reject strong ambient background while improving inverse conditioning, making passive computational periscopy substantially more robust and faster. |
| 2025 | Radar-Based NLoS Pedestrian Localization for Darting-Out Scenarios Near Parked Vehicles with Camera-Assisted Point Cloud Interpretation — Kim et al. | IEEE/RSJ IROS 2025, 21352–21359 | Uses monocular vehicle segmentation/depth to infer temporary parked-vehicle occluders, refines that geometry with 2D mmWave radar points, and then interprets reflected/diffracted returns to localize pedestrians before they emerge into line of sight in measured urban driving experiments. |
| 2026 | Reflection-Aware Reasoning for Non-Line-of-Sight Pedestrian Localization — Park, Jeon, Kim | ECCV 2026 (accepted; project page available before proceedings metadata) | Extends automotive mmWave NLOS localization to a moving ego-vehicle: camera/radar cross-attention predicts reflection type and reflective-surface structure in BEV, then physics-guided ray tracing unfolds distorted multipath. The real outdoor benchmark contains 120 scenarios / 12,539 frames and reports 1.23 m localization accuracy. |
| 2026 | A dual-branch fusion network for footstep sound source localization in non-line-of-sight corridors — Wang, Chen, Yin | The Journal of the Acoustical Society of America 160(2), 1400–1412 (2026) | CorridorLocNet fuses Mel-spectrogram and GCC-PHAT cues with residual-CNN and lightweight-Conformer branches plus cross-attention, learning multipath-dependent footstep position directly from a real around-corner corridor dataset; it reports 98.83% classification accuracy and 2.56 m lower average error than the compared reflection-aware localization baseline. |
| 2026 | Enhanced reflection U-Net reconstruction for passive Non-Line-of-Sight imaging — Yu et al. | Optics and Precision Engineering 34(9), 1496–1506 (2026) | A formally published passive-NLOS U-Net reconstruction study that extends the supervised encoder–decoder branch represented by early passive U-Net/GAN methods and later attention/physics-guided reconstruction. |
| 2025 | Passive Non-Line-of-Sight Imaging via Hyperspectral Autoencoder Net — Li et al. | Proc. SPIE 13542, 135421R (CITA 2024; published 2025) | HAENet extends Hyper-NLOS with spatial–spectral separation reconstruction and spatial–spectral residual attention, exploiting hyperspectral relay measurements while reducing redundancy across wavelength and spatial dimensions. |
| 2024 | Passive non-line-of-sight imaging for hidden target detection — Wang et al. | Proc. SPIE 13282, 132820R (AIIP 2024) | A task-oriented passive-NLOS study focused on hidden-target detection, complementing full hidden-image reconstruction and marking the broader shift toward semantic/passive sensing objectives. |
| 2024 | Reliable reconstruction of passive non-line of sight imaging with occluder by deep learning — Kim, Jang | Proc. SPIE 13076, 1307608 (2024) | Standard-camera occluder-aided passive NLOS reconstruction with deep learning; emphasizes robustness to changes in both occluder and hidden-object position. |
| 2023 | CAGAN: A Channel-aware Generative Adversarial Network for Passive Non-Line-of-Sight Imaging — Yu et al. | IEEE NNICE 2023 | Channel-aware GAN reconstruction on passive NLOS observations, using channel-feature fusion to emphasize informative content and suppress background interference. |
| 2023 | Passive Non-Line-of-Sight imaging reconstruction based on dual input U-Net — He et al. | IEEE NNICE 2023 | Early supervised dual-input U-Net reconstruction for passive NLOS, bridging NLOS-Passive-style data-driven inversion and later attention/generative methods. |
| 2026 | Enhancing passive non-line-of-sight imaging via dynamic channel optimization — Wang et al. | Optics Communications 620, 133626 (2026) | Introduces DCEEM for low-SNR passive NLOS: Bayesian multiplicative fusion combines feature statistics with light-transport descriptors to adaptively suppress noise-dominated channels, while hierarchical denoising/refinement and vector-quantization optimization strengthen hidden-scene reconstruction. |
| 2024 | Hyper-NLOS: hyperspectral passive non-line-of-sight imaging — Chen et al. | Optics Express 32(20), 34807–34824 (2024) | HFN-Net exploits wavelength-resolved relay-wall measurements with a hyperspectral full-color autoencoder and spatial–spectral attention, improving passive-NLOS color/structure recovery and introducing the HS-NLOS dataset. |
| 2025 | Turning rough surfaces into non-line-of-sight cameras — Li et al. | Optica 12(5), 626–634 (2025) | Models microscale rough-wall scattering as an invertible passive encoding rather than a nuisance, enabling ordinary-camera high-resolution real-time NLOS imaging, near-90° field of view, full-color recovery, keyhole imaging, and non-invasive calibration. |
| 2026 | NLOS-MT: A Hybrid Mamba and Windowed Attention Transformer for Non-Line-of-Sight Imaging — Jin et al. | ICPR 2026, LNCS 16816, 297–311 (Springer chapter; first online 3 Aug 2026) | Combines a DeformMamba block for linear-complexity long-range/global transient modeling and robust denoising with a window-attention U-Net for local contextual refinement, extending learned active NLOS reconstruction from pure Transformer or pure Mamba designs to a hybrid state-space/attention architecture. |
| 2023 | Non-line-of-sight imaging and location determination using deep learning — Wang et al. | Optics and Lasers in Engineering 169, 107701 (2023) | Uses a single-shot wall-mediated speckle pattern with SPIR-Net to jointly reconstruct hidden-object appearance and estimate object location, adding spatial localization to passive steady-state NLOS without pulsed ToF or time-gated hardware. |
| 2026 | Symmetry-Aware Gradient Coordination for Physics-Guided Non-Line-of-Sight Imaging — Ling et al. | Symmetry 18(5), 711 (2026) | Treats physics-guided NLOS training as a multi-objective gradient-coordination problem rather than a single weighted scalar loss, combining soft conflict projection, hard physical routing, learnable sensor calibration, and staged optimization for improved low-SNR reconstruction robustness. |
| 2024 | Seeing Around Obstacles Using Active Terahertz Imaging — Cui and Trichopoulos | IEEE Transactions on Terahertz Science and Technology 14(4), 433–445 (2024) | Uses 270–300 GHz active THz imaging and a mirror-folding reconstruction that treats ordinary building surfaces as lossy mirrors, recovering hidden-object geometry and pose with centimeter-scale resolution without prior scene geometry or material properties. |
| 2024 | Around-the-Corner Radar Sensing Using Reconfigurable Intelligent Surface — Yasmeen et al. | IEEE MAPCON 2024 | Demonstrates a custom 1-bit RIS with a 5.5 GHz monostatic radar in a real corridor, electronically redirecting illumination around a corner and recovering hidden-human walking micro-Doppler signatures; the final MAPCON venue supersedes the later arXiv upload. |
| 2025 | Radar Sensing Using Dual-Beam Reconfigurable Intelligent Surface — Yasmeen et al. | IEEE RadarConf25, 1254–1259 (2025) | Studies practical RIS phase quantization for radar sensing: a 1-bit implementation produces dual symmetric beams, and measured/simulated radar cross-section is benchmarked against ideal single-beam RIS and metal-reflector baselines; the final RadarConf25 venue supersedes the later arXiv upload. |
| 2025 | “Around-the-Corner” Radar: Particle Filters for Non-Line-of-Sight Target Tracking in the Presence of Ambiguities — Pham et al. | IEEE Transactions on Aerospace and Electronic Systems 61(3), 5505–5519 (2025) | Formulates multipath around-corner tracking when several propagation hypotheses create ambiguous likelihood modes, and uses particle filtering to maintain the competing target-state hypotheses instead of collapsing onto a wrong trajectory. |
| 2025 | Single-Antenna Non-Line-of-Sight Matrix Imaging via Reconfigurable Intelligent Surfaces — Goïcoechea et al. | arXiv:2512.12359 (2025) | Shows that one transmit–receive antenna plus programmable RIS masks can reconstruct the full reflection matrix, effectively synthesizing an array for high-fidelity NLOS imaging, selective focusing through clutter, and moving-target tracking; no final peer-reviewed venue was verified as of this update. |
| 2025 | Improving SNR for NLoS Target Detection Using Multi-RIS-Assisted Monostatic Radar — Liaquat et al. | IEEE Open Journal of Vehicular Technology 6, 774–789 (2025) | Derives received-power, path-loss, and SNR models for monostatic NLOS radar assisted by multiple RIS relays, quantifying when additional programmable surfaces recover otherwise blocked sensing links and improve target detectability. |
| 2021 | Non-line-of-sight imaging under white-light illumination: a two-step deep learning approach — Zheng et al. | Optics Express 29(24), 40091–40105 (2021) | Introduces an ordinary-camera white-light NLOS system that embeds a speckle-correlation model in a two-stage DNN: the first network regularizes scattered-pattern autocorrelation and the second reconstructs the hidden image, providing the direct precursor to later physics-enhanced white-light NLOS learning. |
| 2026 | Covariance Tensor Decomposition for NLOS Direction Finding in RIS-Aided Bistatic MIMO Radar — Xie et al. | IEEE Signal Processing Letters 33 (2026), 574–578 | Builds a fourth-order covariance tensor for RIS-aided bistatic MIMO radar, extracts the signal subspace with HOSVD, reconstructs steering structure through the Khatri–Rao product, and estimates paired NLOS DOD/DOA efficiently. |
| 2026 | Fast Angle Estimation of NLoS Coherent and Noncoherent Targets via Tensor Decomposition in RIS-Assisted Bistatic MIMO Radar — Yu et al. | IEEE TAES 62 (2026), 8574–8584 | Extends tensor-based RIS-assisted bistatic MIMO radar processing to fast angle estimation for both coherent and noncoherent hidden targets, strengthening the reconfigurable-propagation direction-finding branch. |
| 2025 | Building Corner and NLOS Target Parameter Estimation Based on Diffraction Signal Utilization — Yu et al. | IEEE FUSION 2025, 1–6 | Uses electromagnetic corner-diffraction returns to estimate building-corner and hidden-target parameters, extending diffraction-based UWB NLOS localization beyond the earlier assumption that corner geometry is already known. |
| 2025 | Multipath Ghost Correlation-Based NLOS Target Localization and Building Layuot Estimation — Wei et al. | IEEE EUSIPCO 2025, 2247–2251 | Separates multipath with Range–Doppler features, estimates DOA and ghost positions with IAA, and spatially matches the ghosts to candidate target and wall parameters for joint NLOS localization and building-layout estimation. |
| 2026 | A Cross-Regional NLOS Target Localization Method Based on Joint Multipath GLRT — Yu et al. | IEEE IGARSS 2026, Paper 2579 | A newly presented IGARSS 2026 radar NLOS localization work centered on a joint multipath GLRT for cross-regional target localization; the official program verifies the paper and presentation, while no proceedings DOI was indexed at this update. |
| 2025 | NLoS target localization in IRS-assisted FDA-MIMO radar: A tensor decomposition perspective — Yu et al. | Digital Signal Processing 161 (2025), 105093 | Uses an IRS to establish an NLoS path for bistatic FDA-MIMO radar, estimates target count with sequential MDL, factorizes a third-order echo tensor with PARAFAC, decouples DOD/range and estimates 2-D DOA, then geometrically localizes multiple hidden traffic targets. |
| 2024 | Learning-Based Spotlight Position Optimization for Non-Line-of-Sight Human Localization and Posture Classification — Chandran et al. | IEEE/CVF WACV 2024 | Uses an off-the-shelf projector-camera pair and a message-passing network to infer scene structure and choose the spotlight position that maximizes useful indirect signal, enabling hidden-person localization and posture classification under more arbitrary scene geometry than planar-relay assumptions. |
| 2024 | Deep-Learning-Based Real-Time Passive Non-Line-of-Sight Imaging for Room-Scale Scenes — Li and Zhang | Sensors 24(19), 6480 (2024) | Introduces USEEN for conventional-camera passive NLOS in room-scale scenes, targeting diffuse relay surfaces and ambient-light interference while reporting 12.2 ms inference for real-time hidden-person reconstruction. |
| 2024 | Converting non-confocal measurements into semi-confocal ones with timing-accuracy improving for non-line-of-sight imaging — Zheng et al. | Optics and Lasers in Engineering 176, 108067 (2024) | Transforms point-illumination/SPAD-array non-confocal measurements into semi-confocal histograms with sectionalized-ellipsoid interpolation, using spatial information to improve effective timing accuracy and connect efficient array acquisition to established confocal LCT/f-k reconstruction. |
| 2024 | High-resolution non-confocal non-line-of-sight imaging based on spherical-slice transform from spatial and temporal frequency to space and time — Yu et al. | Optics Letters 49(13), 3806–3809 (2024) | Maps non-confocal transient measurements from spatial/temporal frequency to space/time with a spherical-slice transform, achieving high-resolution reconstruction with reduced artifacts, shape distortion, and position offset; GPU acceleration reduces reconstruction to several hundred milliseconds for a 32×32 PF32 photon-array camera. |
| 2025 | Real-time two-bounce non-line-of-sight object tracking via dual-view collaborative perception network — Zhang et al. | Optics Express 33(20), 42542–42556 (2025) | Uses time-multiplexed alternating dual illumination to acquire two shadow views and DCPNet to fuse cross-view spatial consistency with single-view temporal coherence for real-time hidden-object trajectory estimation; physics-rendered training data and real fine-tuning validate corridor/tunnel-style two-bounce tracking. |
| 2025 | Efficient implicit reconstruction of hidden object in two-bounce non-line-of-sight imaging — Chen et al. | Optics Express 33(19), 41244–41260 (2025) | Replaces explicit voxel carving with hierarchical NeRF-based implicit ray carving and spatial/temporal ray selection, reducing redundant shadow rays while reconstructing static and dynamic hidden scenes; reports about 2% relative depth deviation in a 20 m × 10 m × 4 m volume. |
| 2026 | Material Classification in Acoustic NLOS Environments Using an Attention-Based U-Net and Multimodal Fusion With the ANLOS-R Dataset — Alakuş and Türkoğlu | IEEE Access 14, 26983–27004 (2026) | Introduces ANLOS-R, a 1,440-sample wall-mediated acoustic NLOS dataset collected with an 8-speaker/8-microphone multi-position setup, together with attention-U-Net reflection isolation and multimodal spectral-temporal fusion for hidden-material recognition. It is the dataset/multimodal precursor to the later Sensors 2026 wavelet-feature follow-up. |
| 2026 | Comparative Analysis of Deep Latent Representation and Statistical Fusion Strategies Under Model Inductive Bias in Multichannel Acoustic Live Subject Detection — Cetin et al. | IEEE Access 14, 73343–73356 (2026) | Extends acoustic NLOS semantic sensing from hidden material recognition to live-human versus inanimate-object discrimination in cluttered scenes; compares raw eight-channel measurements, arithmetic averaging, and convolutional-transformer latent fusion across deep and classical classifiers, showing that the best fusion strategy depends strongly on model inductive bias. |
| 2026 | Multipath Exploitation-Based 3-D Environmental Perception and NLOS Moving Target Reconstruction for mmWave MIMO Imaging Radar — Zhu et al. | IEEE Transactions on Aerospace and Electronic Systems 62, 3569–3587 (2026) | Jointly estimates 3-D environmental/reflector structure and reconstructs NLOS moving targets from mmWave MIMO multipath, transferring estimated reflector parameters into path-oriented hidden-target localization and validating the pipeline experimentally. |
| 2025 | NLOS Building Layout and Target Estimation in an L-Shaped Corner with Complex Geometries — Xue et al. | IEEE Transactions on Instrumentation and Measurement 74 (2025) | Removes the usual known-layout assumption: tracks multipath ToAs, separates diffraction and first-/second-order reflection paths, localizes the hidden target and reconstructs part of the L-shaped relay geometry with a portable SISO radar. |
| 2025 | A Two-Stage NLOS Target Positioning Method Based on 0-1 Non-Coherent Binary Accumulation — Wu et al. | IEEE Transactions on Vehicular Technology 74(6), 8866–8878 (2025) | Uses CFAR-derived 0-1 non-coherent binary accumulation and a two-stage estimator to make multipath NLOS target positioning more robust to false alarms and missed detections. |
| 2025 | A Reflective Surface Estimation Method Based on Multipath Utilization — Luo et al. | IEEE Transactions on Instrumentation and Measurement 74, 1–11 (2025) | Estimates the relay/reflective surface itself from multipath-ellipse tangency, dictionary matching and Kalman smoothing, turning a normally assumed calibration quantity into part of the radar inverse problem. |
| 2025 | NLOS Tracking with Distributed Radar Using Multipath-Assisted JPDA — Xu et al. | IEEE IGARSS 2025, 8751–8755 | Extends multipath exploitation from static localization to distributed-radar NLOS tracking with multipath-assisted joint probabilistic data association. |
| 2025 | Non-Line-of-Sight Target Localization in Unknown L-Shaped Corridor Based UWB MIMO Radar — Jia et al. | Journal of Systems Engineering and Electronics 36(3), 681–693 (2025) | Uses dual backprojection views and diffraction/reflection path-length matching to estimate hidden target positions and the unknown corridor width, remaining effective when some propagation paths are lost. |
| 2025 | Non-Line-of-Sight Human Vital-Sign Sensing Aided by Reconfigurable Intelligent Surfaces — Li et al. | Acta Electronica Sinica 53(1), 1–13 (2025) | Uses visual-aided programmable RIS beam control to redirect RF sensing energy into a hidden human region, then estimates respiration and heartbeat with an improved VMD pipeline. This is physiological NLOS sensing rather than hidden-shape reconstruction. |
| 2025 | Liquid Crystal RIS Integrated with SIL Radar for NLOS Vital Sign Monitoring — Tripathy et al. | IEEE IMBioC 2025 | Integrates a liquid-crystal RIS with a self-injection-locked radar, electronically steering the sensing path into an NLOS region for experimentally validated contactless vital-sign monitoring. |
| 2025 | Non-line-of-sight multi-person pose sensing — Hou et al. | Optics Express 33(20), 41937–41950 (2025) | Introduces AMPE-NLOS, the first adaptive multi-person 3D pose/mesh sensing framework for active transient NLOS: LCT supplies coarse volumetric features, a 3D U-Net refines them, and body-center-guided SMPL parameter sampling separates a variable number of hidden people. The paper validates on simulation and a self-built confocal laser/SPAD system. |
| 2024 | Leveraging Rough-Relay-Surface Scattering for Non-Line-of-Sight mmWave Radar Sensing — Xu, Liu, Jiang | IEEE Internet of Things Journal 11(6), 10964–10978 (2024) | Models rough relay walls with stochastic microfacets and turns the resulting scattering diversity into useful NLOS measurements. Its stochastic-geometry-aided three-stage recovery uses virtual ghost targets rather than suppressing them, extending commodity-mmWave sensing beyond ideal smooth-wall specular assumptions. |
| 2024 | Double Sparse Structure-Enhanced mmWave NLOS Imaging Under Multiangle Relay Surface — Xu et al. | IEEE Transactions on Signal Processing 72, 5628–5643 (2024) | Formulates multi-angle-relay mmWave NLOS imaging with an automotive-squint SAR model and a double-sparse hidden-image structure. Time-domain double-sparse thresholding and an approximate frequency-domain operator improve reconstruction under heterogeneous relay orientations in simulation and measured experiments. |
| 2025 | Bayesian Compressive Sensing for NLOS mmWave Imaging Under Imprecisely Multiangle Surfaces — Xu et al. | IEEE Signal Processing Letters 32, 2075–2079 (2025) | Treats relay-surface angles as uncertain dictionary parameters instead of perfectly known geometry. A double-sparse spike-and-slab prior, expectation-propagation inference, and EM-based parameter updates reduce NLOS reconstruction error when multi-angle relay layouts are imprecisely calibrated. |
| 2025 | mmWave-Based Relay Reflector Reconstruction for LiDAR-Free Around-Corner Human Sensing — Lv et al. | IEEE INFOCOM 2025, 1–10 | Reconstructs the relay reflector from mmWave measurements themselves before around-corner human sensing, removing the common dependence on LiDAR or manually supplied wall geometry. This makes reflector geometry part of the sensing problem rather than a fixed calibration input. |
| 2024 | Hydra: Exploiting Multi-Bounce Scattering for Beyond-Field-of-View mmWave Radar — Mehrotra et al. | ACM MobiCom 2024, 1545–1559 | Explicitly models and exploits multi-bounce scattering so a standalone commodity mmWave radar can localize objects outside its transmit beam, behind occlusions, or even behind the radar without prior environment knowledge. Across five real environments it reports 2×–10× lower median beyond-FOV localization error than baselines. |
| 2026 | MmWave NLOS Sensing under Rough Relay Surface: Challenges and Solutions — Liu et al. | IEEE Aerospace and Electronic Systems Magazine, 2026 | Synthesizes the emerging rough-relay mmWave NLOS problem: real building materials and non-flat surfaces break smooth-specular assumptions but also create exploitable scattering diversity. The article organizes the modeling, multipath-management, and recovery challenges that connect rough-wall sensing to the 2024–2025 reconstruction methods. |
| 2025 | Non-line-of-sight sound source localization based on block sparse Bayesian learning and second-order edge diffraction — Zhai et al. | Applied Acoustics 228, 110369 (2025) | Uses the Biot–Tolstoy–Medwin second-order edge-diffraction response as a sensing operator and block sparse Bayesian learning to localize hidden acoustic sources when direct and first-order paths are unavailable; validated with simulation and a 32-channel array. |
| 2025 | Non-Line-of-Sight Multi-Target Localization in T-Junctions Using Ray Tracing of mmWave Radar — Jeon et al. | IEEE Intelligent Vehicles Symposium (IV), 1779–1786 (2025) | Infers T-junction layout from static 2D mmWave points, ray-traces dynamic multipath returns, and filters/clusters the unfolded points to localize multiple hidden pedestrians in measured outdoor experiments. |
| 2025 | BiScalar-AA: BiScalar Attentive Amplifier Network for NLOS Object Detection and Tracking Using Millimeter-Wave Radar — Yu et al. | IEEE Smart World Congress (SWC), 886–893 (2025) | Maps sparse mmWave point clouds to pseudo-images and applies an attentive amplification network for hidden-object detection and tracking in dynamic outdoor scenes, extending radar NLOS toward learned semantic perception. |
| 2025 | Two-Stage Attention Network for NLOS Object Detection and Tracking Using mmWave Radar — Yu et al. | Computer Engineering, online first 9 September 2025 | Uses a two-stage attention architecture on radar pseudo-images for real-time hidden-target detection and tracking, reporting 75.62% mAP and a 5.99-point improvement over the compared baseline. |
| 2022 | CornerRadar: RF-Based Indoor Localization Around Corners — Yue et al. | PACM IMWUT 6(1), Article 34 (2022) | Introduces a learned RF hint-map representation that captures practical indoor multipath and localizes people around corners across 56 environments, reducing median error by 3–12× versus prior RF baselines. |
| 2022 | Mosaic: Leveraging Diverse Reflector Geometries for Omnidirectional Around-Corner Automotive Radar — Woodford et al. | ACM MobiSys, 155–167 (2022) | Uses multiple curved and planar environmental reflectors to enlarge around-corner automotive-radar coverage, more than doubling hidden-vehicle detection probability at three real urban sites. |
| 2024 | Around the Corner mmWave Imaging in Practical Environments — Dodds et al. | ACM MobiCom, 953–967 (2024) | RFlect exploits practical environmental reflectors, including poles and curved/composite surfaces, to reconstruct hidden object shape beyond the planar-wall assumptions of earlier around-corner radar systems. |
| 2025 | Non-Line-of-Sight 3D Object Reconstruction via mmWave Surface Normal Estimation — Dodds et al. | ACM MobiSys, 445–458 (2025) | mmNorm estimates surface-normal fields from mmWave observations and integrates them into hidden-object geometry, moving RF NLOS from localization and coarse reflectivity toward full 3D shape reconstruction. |
| 2026 | Wave-Former: Through-Occlusion 3D Reconstruction via Wireless Shape Completion — Dodds et al. | CVPR, 21713–21724 (2026) | A physics-aware transformer bridges raw mmWave sensing with wireless shape completion, using candidate surfaces, learned completion, and entropy-guided selection to recover complete 3D geometry of fully occluded everyday objects. |
| 2026 | RISE: Single Static Radar-based Indoor Scene Understanding — Zhou et al. | CVPR, 32194–32205 (2026) | Uses AoA/AoD multipath enhancement plus sim-to-real hierarchical diffusion for layout reconstruction and object detection from one static radar, establishing a large-scale benchmark for radar-only indoor scene understanding. |
| 2025 | Rapid Indoor Mapping and Non-Line-of-Sight Imaging Using a 228-GHz FMCW Polarimetric Radar System — Alburadi et al. | IEEE Transactions on Geoscience and Remote Sensing 63, 1–9 (2025) | A 222–228 GHz polarimetric FMCW radar with a mechanically scanned fan beam produces high-resolution indoor maps and exploits double-bounce specular paths for around-corner detection; polarization diversity helps separate useful NLOS returns from leakage and clutter. |
| 2026 | Material Classification from Non-Line-of-Sight Acoustic Echoes Using Wavelet-Acoustic Hybrid Feature Fusion — Alakuş and Türkoğlu | Sensors 26(5), 1577 (2026) | Uses chirp echoes reaching nine hidden materials only through wall-mediated acoustic paths. A 70-dimensional fusion of classical acoustic and multi-scale wavelet features, classified by recurrent networks, reaches 0.99 balanced accuracy and macro-F1; SHAP links predictions to interpretable material properties. This is semantic acoustic NLOS sensing rather than hidden geometry reconstruction. |
| 2025 | Non-line-of-sight virtual modulated range migration imaging based on super-resolution histograms — Tian et al. | Optics Letters 50(2), 519–522 (2025) | Combines deconvolution-modified iterative backprojection with virtual modulated range migration, recovering 50× super-resolved histograms from 1 ns measurements before high-resolution confocal or non-confocal reconstruction. It targets inexpensive low-time-resolution NLOS systems and reduces the required data volume. |
| 2022 | Fast non-line-of-sight imaging based on product-convolution expansions — Xu et al. | Optics Letters 47(18), 4680–4683 (2022) | Approximates the shift-variant non-confocal ellipsoidal forward operator and its adjoint by local convolutions, then uses FFTs and low-rank matrix decompositions for fast iterative reconstruction. Public transient datasets show phasor-field-like quality with substantially lower runtime. |
| 2026 | Non-line-of-sight vehicle detection via audio-visual fusion — Wang et al. | IEEE ICASSP 2026, 11807–11811 | Uses bird's-eye-view scene geometry together with time-frequency and spatiotemporal acoustic spectra in a CNN–LSTM–Conformer network for detecting fully occluded vehicles. The published experiments report 94.1% and 97.0% accuracy on the OVAD and AOVD datasets, extending acoustic NLOS from localization and reconstruction toward scene-conditioned semantic perception. |
| 2025 | Localizing acoustic sources in non-line-of-sight scenarios using irregular-grid beamforming and first-order edge diffraction — Zhai et al. | Measurement 256, 117944 (2025) | Builds a non-free-field steering vector from the Biot–Tolstoy–Medwin first-order edge-diffraction response and combines it with irregular-grid frequency-domain beamforming. Simulations and a 32-channel array experiment localize hidden acoustic sources while limiting main-lobe deformation behind finite obstacle edges. |
| 2025 | Non-Line-of-Sight Vehicle Localization Based on Sound — Jeon et al. | IEEE Transactions on Intelligent Transportation Systems 26(2), 2321–2338 (2025) | Introduces an Acoustic-Spatial Pseudo-Likelihood particle filter for tracking vehicles approaching from fully occluded road regions. ARIL supplies BEV ground-truth positions and, together with OVAD, validates a practical passive acoustic NLOS localization branch for collision avoidance. |
| 2025 | Contrast Adaptive Slot-Attention Network for NLoS Classification Under Temporal Truncation — Lin et al. | IEEE EICARS 2025, 140–143 | Uses CLAHE/Sobel-guided contrast modulation and physically guided slot regularization to stabilize object-centric attention on temporally truncated, photon-sparse NLOS signals. It extends the learned-transient branch from reconstruction toward robust semantic classification under incomplete time-series measurements. |
| 2025 | Enhanced passive non-line-of-sight imaging via multi-scale polarization-guided diffusion model — Jin et al. | The Visual Computer 41(13), 10789–10804 (2025) | MSPDiff progressively reconstructs polarized long-wave-infrared relay observations from coarse to fine resolution, using polarization as a physical cue inside diffusion sampling; the reported passive dataset results reach 25.78 dB PSNR and 0.92 SSIM. |
| 2025 | Lightweight multi-scale feature fusion with attention guidance for passive non-line-of-sight imaging — Chen et al. | The Visual Computer 41(10), 7767–7780 (2025) | LMS-NLOS combines multi-scale encoding, detail-enhanced Transformer blocks, asymmetric shallow/deep fusion, contour-aware loss, and spatial-shift feed-forward units; its lightweight variant nearly halves model size while retaining strong passive reconstruction quality. |
| 2025 | Multipath Exploitation-Based Linearized Inverse Scattering Method for Non-Line-of-Sight Indoor Imaging of PEC Objects — Suenobu et al. | IEEE JSTARS 18, 6694–6709 (2025) | Treats wall-mediated multipath as a geometry-specific numerical Green tensor inside a physical-optics linearized inverse-scattering model. Full-wave simulation and measured 2 GHz T-junction experiments recover the location and planar extent of hidden PEC objects. |
| 2024 | Non-Line-of-Sight Imaging by Linearized Inverse Scattering Method Based on Physical Optics — Suenobu et al. | IEEE ICEAA 2024 | Establishes the simulation-only precursor: known T-junction geometry and a numerically computed multipath Green tensor make physical-optics linearized inverse scattering applicable to hidden PEC targets. |
| 2024 | Enhancing the spatial resolution of time-of-flight based non-line-of-sight imaging via instrument response function deconvolution — Wang et al. | Optics Express 32(7), 12303–12317 (2024) | Models measured transients as a Poisson convolution with the calibrated instrument response and deconvolves timing blur before LCT or f-k migration. Simulations and experiments show that reconstruction remains viable with total timing jitter up to 1200 ps, linking detector calibration directly to recoverable spatial resolution. |
| 2023 | Compressive Non-Line-of-Sight Imaging with Deep Learning — Zhu et al. | Physical Review Applied 19(3), 034090 (2023) | Reconstructs 32×32 hidden images from only 8×8 transient scan points using a CNN trained entirely on simulation and transferred to measured data. The 6.25% spatial sampling regime is an early direct precursor to later learned under-scanning and transient-completion methods. |
| 2022 | Non-line-of-sight imaging based on an untrained deep decoder network — Wu et al. | Optics Letters 47(19), 5056–5059 (2022) | Couples an untrained deep decoder with the passive occluder-aided forward model, optimizing network weights per measurement without paired training data. It improves hidden-image detail and robustness under strong ambient light, bridging computational periscopy and zero-shot physics-guided neural priors. |
| 2024 | Multi-wavelength single-pixel non-line-of-sight imaging with a compressive sensing measurement matrix — Li et al. | Applied Physics B 130(7), 127 (2024) | Extends patterned single-pixel NLOS from monochrome spatial recovery to RGB hidden-scene imaging. Multi-wavelength measurements and TV-regularized compressive reconstruction recover colour and spatial detail at a reported 29% compression rate. |
| 2023 | Active mode single-pixel non-line-of-sight imaging system based on second-order correlation and diffraction — Li et al. | Journal of Optics 25(1), 015702 (2023) | Combines patterned active illumination and second-order correlation with Fourier-domain inverse diffraction. Hadamard coding improves measured hidden-image recovery and reaches a reported SSIM of 0.9286 at full sampling. |
| 2019 | 3D RGB Non-Line-of-Sight single-pixel imaging — Musarra et al. | Imaging Systems and Applications / OSA Technical Digest 2019, IM2B.5 | Demonstrates sub-second full-colour 3D hidden-scene acquisition with a high-efficiency time-resolved single-pixel camera, establishing an early bridge between transient depth and RGB single-pixel NLOS. |
| 2025 | Shoot-Bounce-3D: Single-Shot Occlusion-Aware 3D from Lidar by Decomposing Two-Bounce Light — Klinghoffer et al. | SIGGRAPH Asia 2025 | Uses multiplexed single-shot single-photon LiDAR and a learned decomposition of two-bounce transport, trained on roughly 100,000 simulated transients, to recover dense depth plus occluded and mirror-mediated geometry from one view. Real captures validate the demultiplexing; this is tightly adjacent multi-bounce NLOS reconstruction rather than the conventional relay-wall configuration. |
| 2025 | Transient LASSO: Transient Large-Scale Scene Reconstruction — Scheuble et al. | SIGGRAPH Asia 2025 | Fits a neural scene representation directly to posed raw outdoor transient-LiDAR measurements while explicitly modeling back-reflected light, ambient illumination, and high-flux sensor behavior. It disentangles dense geometry and normals from reflectivity, retroreflectivity, and ambient light, extending transient inverse rendering to in-the-wild urban scenes. |
| 2026 | Real-Time and High-Fidelity Non-Line-of-Sight Imaging — Ji et al. | Research Square preprint 2026 | Introduces a unified inverse framework for both see-through-scattering-media and see-around-corner NLOS. Scale modulation and joint regularization recover hidden albedo and depth across diverse measurement settings; no final journal or conference venue was verified. |
| 2025 | HOLI-1-to-3: Transient-Enhanced Holistic Image-to-3D Generation — Shen et al. | IEEE TPAMI 47(9), 7206–7217 (2025) | Unifies LOS radiance fields and NLOS transient fields in a neural plenoptic representation; diffusion and transient priors recover both visible and invisible object geometry from a single viewpoint. |
| 2025 | Vectorial Digitelligent Optics for High-Resolution Non-Line-of-Sight Imaging — Guo et al. | Engineering 45, 70–78 (2025) | Jointly optimizes wavefront phase and polarization through a scattering relay wall, then raster-scans the optimized focus inside the optical-memory-effect range. The measured system reaches 0.40 mm resolution at 0.35 m, moving part of high-resolution NLOS inversion into closed-loop physical illumination control. |
| 2024 | Vectorial-Optics-Enabled Multi-View Non-Line-of-Sight Imaging with High Signal-to-Noise Ratio — Wang et al. | Laser & Photonics Reviews 18(6), 2300909 (2024) | Uses a vector-optical relay-reflection model to select illumination angle plus incident and received polarization, creating complementary high-SNR views for hidden-object reconstruction and recognition under weak returns. |
| 2024 | High-Resolution Non-Line-of-Sight Imaging Based on Liquid Crystal Planar Optical Elements — Zhao et al. | Nanophotonics 13(12), 2161–2172 (2024) | Inserts a liquid-crystal planar angle magnifier to enlarge the effective transient relay-wall aperture; measurement-correlation-aware sparse scanning reduces acquisition time by more than 20% while retaining the demonstrated resolution. |
| 2026 | Early fusion of laser and acoustic features for human orientation detection in non-line-of-sight environments — Doğan | Scientific Reports 2026 | Fuses 21 laser-chirp and 21 acoustic-chirp features into a 42-feature representation and evaluates conventional classifiers, LAO-Net, and explainable-AI analysis for four hidden-person orientations in controlled NLOS experiments. This is multimodal semantic NLOS sensing, not hidden-image or 3D-geometry reconstruction. |
| 2025 | Physics to the Rescue: Deep Non-Line-of-Sight Reconstruction for High-Speed Imaging — Mu et al. | IEEE TPAMI 2025 | Embeds complementary wave-propagation and volume-rendering priors in a feed-forward model for high-speed non-confocal transient NLOS. The model is trained on synthetic data with intensity or raw-transient supervision, generalizes to measured captures, and reconstructs intensity and depth at more than 5 captures per second. |
| 2025 | Reconfigurable intelligent surface-enabled gridless DoA estimation system for NLoS scenarios — Yuan et al. | Signal Processing 2025 | Uses an RIS-created virtual-LOS path, covariance-domain denoising, atomic-norm minimization, and ADMM for gridless multi-target direction estimation with limited receive hardware; numerical validation only. |
| 2026 | RIS-aided monostatic radar for NLOS target DOA estimation based on steering vector decoupling — Zhang et al. | Signal Processing 2026 | Scans hidden directions with an RIS codebook, decouples the target steering-vector outer product from the composite monostatic echo, and applies Root-MUSIC for NLOS angle estimation; simulation-only, not hidden-shape reconstruction. |
| 2024 | Multispectral imaging through scattering media and around corners via spectral component separation — Wei et al. | Optics Express 2024 | Builds a multispectral-speckle simplex from wavelength-decorrelated components and combines spectral intensity modulation, wavelength-count estimation, vertex-component initialization, constrained non-negative matrix factorization, and phase retrieval. Experiments recover up to six spectral channels and validate multispectral hidden-object imaging around a diffuse corner. |
| 2025 | Single-shot multitarget imaging beyond the OME range around the corner through polarization component extraction — Zhao et al. | Proc. SPIE / OYSS 2025 | Introduces polarization-component extraction for one-shot recovery of multiple around-corner targets located in distinct optical-memory-effect regions. This short conference precursor establishes the polarization-separation idea later generalized into the 2026 spatial-multiplexing journal system. |
| 2026 | 3D Reconstruction of Hidden Objects from Simultaneous Recovery of Light Source and Environment — Matsubara et al. | VISAPP 2026 | Uses far-infrared reflections from walls and floors to reconstruct moving hidden luminous objects while jointly estimating unknown scene parameters. Self-supervised reprojection on real measurements is combined with supervised synthetic training, extending passive thermal NLOS from known calibrated environments toward joint scene-and-target recovery. |
| 2026 | Estimating the 3D Position of Hidden Humans Using Reflections on Vehicle Bodies — Kozawa et al. | VISAPP 2026 | Treats curved vehicle bodies as opportunistic convex mirrors: a video detector finds distorted pedestrian reflections, while monocular depth, surface normals, and a specular-reflection loss recover the hidden pedestrian's 3D road position. This is passive NLOS localization rather than full hidden-scene reconstruction. |
| 2020 | NLOS Obstacle Position Estimation from Reflected Image — Takatori | IEEE IV 2020 | Uses stereo observations of virtual road-obstacle images reflected by nearby vehicle bodies or roadside glass and geometrically unfolds the reflection to estimate hidden-obstacle position; a 1/5-scale study establishes the foundational automotive passive/specular NLOS localization formulation. |
| 2022 | Estimation of NLOS Obstacle Position Using Reflected Image on Transparent Surface — Nakamura, Takatori | IEEE ITSC 2022 | Extends stereo reflected-image NLOS localization to transparent relay surfaces, addressing parallax, incidence angle, and polarization-filter effects; this remains task-oriented obstacle localization rather than hidden-scene reconstruction. |
| 2025 | Object Detection Method for Non-Line-of-Sight Obstacles Reflected on Painted Surfaces — Oyama, Takatori | IEEE ITSC 2025 | Moves the automotive reflection lineage from geometric localization to semantic detection on weak, colour-distorted painted-surface reflections; direct-view normalization followed by a general pretrained detector reports 85.7% overall accuracy. |
| 2026 | Feasibility study of non-line-of-sight obstacle location estimation using reflected images from in-vehicle sensors — Sheng, Takatori | IATSS Research 2026 | Combines real-road reflection-frequency observations, microscopic traffic simulation, reflective-surface-angle modeling, and stereo virtual-image unfolding; 1/6-scale experiments report 4.3 cm mean error at 1.7–3.7 m and approximately 45 mm RMS error under reflector-orientation changes. |
| 2026 | Single-shot imaging through scattering media and around the corner beyond the OME range via polarization-encoded spatial multiplexing — Wei et al. | Optics and Lasers in Engineering 2026 | Polarization-encoded spatial multiplexing demultiplexes speckles from distinct optical-memory-effect regions in one exposure, enabling multi-target hidden-scene recovery beyond the conventional memory-effect field of view; experiments report 32 dB demultiplexing fidelity and 3.75× field-of-view expansion. |
| 2026 | Non-Line-of-Sight Perception Method for Autonomous Haul Trucks in Open-Pit Mines Based on 4D mmWave Radar and LiDAR Fusion — Yang et al. | Sensors 2026 | BSCF geometrically filters multipath-corrupted 4D radar and selectively injects high-confidence radar points into 3D-LiDAR blind regions in real open-pit-mine data. It provides hidden-target existence and envelope-level risk cues rather than full semantic detection or 3D reconstruction. |
| 2026 | Fast SPAD-array timing-error correction with time-referencing for non-line-of-sight imaging | Optics Express | Corrects approximately 6% pixel-to-pixel mean-bin-width variation and 0.6% bin-level fluctuation in SPAD-array TCSPC data with a precomputed lookup table, then derives an absolute NLOS time reference from intrinsic diffraction-wave and lens-flare peaks; the calibration sharpens measured non-confocal and simulated confocal reconstructions without an external timing target. |
| 2026 | Fast NLOS imaging at low sampling rates based on Poisson modeling | Proc. SPIE / AOMATT 2025 | Formulates photon-limited transient inversion with a Poisson likelihood and a GPU-parallel Nesterov proximal-gradient solver; measured 64×64×4096 transients retain strong reconstruction quality at 1.56% sampling with roughly one-second processing per frame. |
| 2026 | Semantic-guided under sampling scanning for improving real-time performance of NLOS imaging | Proc. SPIE / CITA 2025 | Uses a semantic target contour to restrict the relay-wall scan to one quarter of the original area and then applies neighborhood sparse sampling, reducing density to one fifth while limiting the reported PSNR and SSIM degradation. |
| 2026 | Non-Line-of-Sight Target Recognition Method Based on Multi-Scale Feature Fusion — Zeng et al. | Journal of Signal Processing 2026 | Uses a measured 15 GHz stepped-frequency radar, adaptive multi-scale convolutions and attention to fuse salient multipath and local-scattering cues, classifying four hidden target types with 99.6% accuracy; this is semantic recognition rather than hidden-shape reconstruction. |
| 2026 | Multipath Contrastive Learning for Non-line-of-sight Human Activity Recognition Using an Ultrawideband Radar — Zhong et al. | Journal of Radars 2026 (Online First) | MuPhyCoNet treats separated multipath time-frequency views as natural contrastive positives and adds observation/prediction physics constraints; on 19,500 measured UWB-radar spectrograms it reaches 94.32% six-action accuracy with only 10% labels. |
| 2024 | Speckle-correlation-based non-line-of-sight imaging under white-light illumination — Zhou et al. | Optics & Laser Technology 2024 | ZPF-SCI combines Zernike-polynomial envelope correction with low-pass filtering before speckle autocorrelation and phase retrieval, retaining reconstruction under ambient illumination and camera misalignment with an ordinary camera. |
| 2025 | Non-line-of-sight imaging under white-light illumination using physics-enhanced deep learning — Fu et al. | Applied Optics 2025 | Embeds a speckle-correlation forward model and a denoising prior in a physics-enhanced network, enabling ordinary-camera hidden-image reconstruction under inexpensive broadband white-light illumination. |
| 2025 | Single-shot non-line-of-sight imaging based on the statistical average characteristics of a speckle pattern under ambient light — Zhou et al. | Optics Communications 2025 | Extracts object-spectrum amplitude from the covariance of one random speckle pattern, removing multi-frame stitching and the conventional memory-effect field-of-view limit while reconstructing at −2.06 dB SNR. |
| 2025 | Isolating Signals in Passive Non-Line-of-Sight Imaging Using Spectral Content — Hashemi et al. | IEEE TPAMI 2025 | Uses multispectral unmixing and a convex known-spectrum formulation to separate desired wall-mediated radiance from much stronger clutter, improving passive reconstruction under realistic backgrounds. |
| 2025 | Hyperspectral passive non-line-of-sight imaging with band selection — Chen et al. | Expert Systems with Applications 2025 | HSBS-Net selects informative spectral bands, applies a spectral-energy-guided KA-Transformer and robust sparse loss, and introduces HP-NLOS for physical full-colour passive hyperspectral reconstruction. |
| 2025 | CMFormer: Non-line-of-sight imaging with a memory-efficient MetaFormer network — Zhang et al. | Optics and Lasers in Engineering 2025 | A convolutional MetaFormer token mixer, aggregate feature transmission, cross-layer attention and checkpointing reduce 3D transient memory cost while reaching 8-fps reconstruction on consumer GPUs. |
| 2025 | Time-multiplexing non-line-of-sight imaging — Li et al. | Chinese Optics Letters 2025 | Introduces delayed multi-beam temporal multiplexing so one time-resolved channel separates echoes from multiple relay points; the proof-of-concept system reconstructs hidden scenes with half the conventional relay scans. |
| 2024 | Non-Line-of-Sight Target Tracking With a Single Time Multiplexed Channel — Zheng et al. | IEEE Photonics Journal 2024 | Uses two delayed illumination paths to encode multiple hidden-target echoes into one single-pixel single-photon histogram, enabling centimeter-precision positioning and tracking with a single time-correlated detection channel. |
| 2024 | Non-line-of-sight target 2D size detection with few channels at a time — Li et al. | Expert Systems with Applications 2024 | Learns a rectangle-based mapping from only two or three transient detection channels to hidden-target width and height, extending scan-free few-channel NLOS from localization to coarse 2D size inference rather than full scene reconstruction. |
| 2026 | Eye-Safe Non-Line-of-Sight Localization Using Compact Nanosecond Laser Diodes and Single-Photon-Avalanche-Diode Arrays — Albert et al. | JEOS-RP 2026 | Combines inexpensive nanosecond laser diodes with a parallel SPAD array, dual off-axis illumination, matched temporal filtering, and ellipsoidal backprojection for compact eye-safe hidden-target localization. |
| 2026 | SCISA-Net: Scene-Constrained Inverse-to-Subband Attention for Semantic Inference from Wall-Mediated Indirect Observations — Dai et al. | Photonics 2026 | Scene-constrained regularized inversion and multi-stage Haar-subband attention infer 31 hidden-display categories from calibrated wall-mediated observations; the task is semantic NLOS sensing rather than complete image or geometry reconstruction. |
| 2026 | Dual-model guided active NLOS imaging with under-scanning measurements — Yan et al. | The Visual Computer 2026 | A spatio-temporal recovery module completes sparse transients before dual LCT and f-k branches recover complementary global structure and fine texture, with adaptive feature fusion for efficient under-scanned reconstruction. |
| 2026 | Non-line-of-sight imaging based on adaptive neural grid resampling — Wang et al. | SSRN preprint 2026 | A lightweight Grid Offset Network predicts local frequency-domain sampling offsets and adaptively calibrates the f-k Stolt mapping under timing, propagation-speed, and sparse-sampling perturbations while preserving fast migration. |
| 2022 | Nonline-of-Sight 3-D Imaging Using Millimeter-Wave Radar — Wei et al. | IEEE TGRS 2022 | Establishes a measured 77 GHz MIMO around-corner 3D imaging model, resolution analysis, mirror-symmetry backprojection, and relay-surface phase-error compensation, providing the physical foundation for later model-driven mmWave NLOS reconstruction. |
| 2024 | Precise Reconstruction Method for Hidden Targets Based on Non-line-of-sight Radar 3D Imaging — Cai et al. | Journal of Radars 2024 | Separates LOS and NLOS multipath echoes and jointly estimates hidden reflectivity, total-variation structure, and relay-surface phase error; measured corner experiments reconstruct knives and metal racks under aperture occlusion. |
| 2024 | RM-CSTV: An Effective High-Resolution Method of Non-Line-of-Sight Millimeter-Wave Radar 3-D Imaging — Liu et al. | National Science Open 2024 | Couples a fast range-migration kernel with complex sparsity and total variation, corrects virtual-target geometry, and preserves hidden contours under sparse sampling while reducing computation by roughly two orders of magnitude versus matrix-based CSTV. |
| 2025 | Non-Line-of-Sight 3D Reconstruction with Radar (HoloRadar) — Lai, Lan, Zhao | NeurIPS 2025 | Uses a single mobile mmWave radar to predict multi-return range images and then map mirrored observations to their physical locations with a ray-interaction-aware network, reconstructing complete LOS and hidden 3D scenes in real environments. |
| 2025 | Non-Line-of-Sight mmW Radar Imaging With Adaptive Artifact Cancellation and Target Enhancement — Cai et al. | IEEE TAP 2025 | Introduces unsupervised ACTE-Net: Gaussian-convolution artifact cancellation, energy-based target enhancement, and a 2D holographic operator suppress multipath blur without storing a large sensing matrix; measured hidden-letter and trellis experiments validate 3D reconstruction. |
| 2025 | Non-Line-of-Sight mmW SAR Imaging With Equivariant Adaptive Threshold Learning — Cai et al. | IEEE TCI 2025 | Proposes a fully self-supervised equivariant-imaging framework with TV-constrained deep unfolding and an adaptive peak-convolution threshold module, reconstructing measured 2D/3D mmWave SAR images from partial, noisy, multipath-corrupted NLOS echoes. |
| 2026 | RM Operator Learning-driven Non-line-of-sight 3D Imaging Method for Millimeter Wave Radar — Chen et al. | Journal of Radars 2026 | Embeds a fast range-migration operator in a FISTA-derived unfolding network to compensate phase error, aperture shadowing, and multipath; measured near-field targets under ideal and non-ideal reflectors show higher precision and about 100× speedup over sparse solvers. |
| 2026 | Learning to See Around Corners: A Deep Unfolding Framework for Terahertz Radar Non-Line-of-Sight 3D Imaging — Chen et al. | Photonics 2026 | Extends model-driven NLOS radar reconstruction to a measured 121 GHz platform, embedding a fast holographic operator in FISTA-Net to correct phase error, aperture occlusion, and multipath while avoiding a prohibitively large sensing matrix. |
| 2026 | Iterating the transient light transport matrix for non-line-of-sight imaging — Sultan et al. | Nature Communications 2026 | Measures the full first-order transient light-transport matrix with dense relay-wall illumination and a gated 16×16 SPAD array, then independently focuses virtual illumination and detection inside the hidden scene to recover hidden-to-hidden TLTM-2. The recovered transport exposes indirect shadows, interreflections, and volumetric scattering and supports direct/indirect separation, virtual relighting, and dual photography. |
| 2026 | Ptychography-enhanced correlography for non-line-of-sight imaging — Liu et al. | Advanced Photonics Nexus 2026 | Introduces PEC-NLOS, which scans overlapping indirect speckle probes across the relay wall and applies ptychographic phase retrieval to jointly refine the hidden object and probe. A geometry-based 3D data-correction stage compensates scan-position and probe distortion, extending steady-state correlography from simple targets to high-spatial-frequency, high-entropy hidden scenes. |
| 2025 | High-resolution non-line-of-sight imaging via polarization differential correlography — Liu et al. | Chinese Optics Letters 2025 | Introduces PDC-NLOS, a single-shot steady-state system that encodes hidden objects with independently polarized laser speckles and uses polarization differencing before correlographic phase retrieval. It removes mechanical relay scanning, improves perturbation robustness, and demonstrates millimeter-level resolution with an average reported SSIM of about 0.76. |
| 2025 | A hybrid perceptron with cross-domain transferability towards active steady-state non-line-of-sight imaging — Liang et al. | Signal Processing 2025 | Introduces HP-CDT for active steady-state NLOS. Hierarchical pooling and feature fusion combine local CNN and global Transformer perception, while LOS-domain latent representations guide NLOS feature learning; synthetic and measured experiments demonstrate lightweight, high-fidelity 2D hidden-scene reconstruction. |
| 2026 | D-NeSF: Dynamic Neural Shadow Fields for Two-bounce Non-line-of-sight Stereo Reconstruction — Zhang et al. | ICGIP 2025 / Proceedings of SPIE 14124 (2026) | Extends two-bounce shadow-based NLOS from static occupancy or illumination fields to moving hidden geometry. A spatiotemporally disentangled six-plane neural field, cyclic multi-position illumination, and self-supervised cumulative-transmittance constraints reconstruct temporally consistent dynamic 3D targets from shadow sequences. |
| 2026 | Detection of sUAS in Urban Environments using Multi-Antenna Micro-Doppler Radar — Liyanage et al. | IEEE VTC2026-Fall (accepted) | Uses a hardware-validated 2.47 GHz four-receiver CW MIMO radar and cyclostationary propeller micro-Doppler features for hidden-drone detection in indoor and semi-urban NLOS settings. A compact EfficientNet-B0 model reaches 86.11% overall accuracy; the result is detection-only, with stationary airframes and no hidden-target localization or 3D reconstruction. |
| 2022 | Fast non-line-of-sight imaging based on first photon event stamping — Li et al. | Optics Letters 2022 | Introduces time-sequential first-photon (TSFP) acquisition for active transient NLOS, modeling the detection process rather than changing the downstream inverse operator. Synthetic and measured experiments retain comparable reconstruction quality with substantially shorter acquisition, making the method relevant to photon-starved and real-time systems. |
| 2026 | Deep Learning–Aided Frequency-Modulated Continuous-Wave Radar for Around-the-Corner Non-Line-of-Sight Perception at Urban Intersections — Lin, Chen | Computer Modeling in Engineering & Sciences 2026 | Uses simulated 77 GHz automotive FMCW echoes and a compact residual AlexNet-derived chirp-restoration network to exploit building specular reflections for around-corner range and angle estimation at urban intersections. The paper is an application-facing NLOS radar-perception result; all reported evaluations are simulation-only, with measured-data validation left for future work. |
| 2024 | Non-Line-of-Sight Imaging and Vibrometry Using a Comb-Calibrated Coherent Sensor — Huang et al. | Physical Review Letters 2024 | Establishes coherent optical NLOS sensing with an optical-frequency-comb-calibrated FMCW LiDAR, providing sub-picosecond effective temporal resolution, submillimeter hidden-scene localization and 3D imaging, and frequency-resolved NLOS vibrometry under strong ambient light. |
| 2025 | High-Resolution Non-Line-of-Sight Tracking by Comb-Calibrated FMCW LiDAR — Ye et al. | Laser & Photonics Reviews 2025 | Extends comb-calibrated coherent NLOS from static imaging to snapshot multi-object tracking, reporting 2 mm 3D position accuracy and 2 mm/s velocity accuracy without temporal accumulation. |
| 2025 | Non-Line-of-Sight Ranging and 3D Imaging Using Vector Enhanced Sensitive FMCW LiDAR — Chen et al. | Journal of Lightwave Technology 2025 | Uses laser-feedback interferometry, polarization-vector enhancement, and K-domain resampling for a simpler sensitive FMCW architecture, demonstrating better-than-32-µm NLOS absolute ranging and millimeter-level hidden 3D imaging at about 2.8 m. |
| 2024 | Efficient non-line-of-sight tracking with computational neuromorphic imaging — Zhu et al. | Optics Letters 2024 | Uses an event camera to retain only motion-induced changes in dynamic NLOS speckle, suppressing static relay/background redundancy and enabling direct, efficient hidden-target motion estimation without first reconstructing a complete hidden scene. |
| 2025 | Zero-Phase Phasor Fields for Non-Line-of-Sight Imaging — Luesia-Lahoz et al. | IEEE ICCP 2025 | Uses zero crossings of the virtual phasor-field phase as a geometry cue, obtaining highly precise hidden depth while retaining the noise robustness of phase-based wave propagation. |
| 2024 | Towards a More Accurate Light Transport Model for Non-Line-of-Sight Imaging — Sultan, Reza, Velten | Optics Express 2024 | Derives a more complete transient transport model that unifies time-of-flight and shadow/occlusion NLOS formulations and exposes approximations made by widely used three-bounce models. |
| 2024 | Cohesive Framework for Non-Line-of-Sight Imaging Based on Dirac Notation — Redo-Sanchez et al. | Optics Express 2024 | Places backprojection, f-k migration, and phasor-field propagation in one operator notation, clarifying their mathematical relationships and the role of Rayleigh–Sommerfeld propagation. |
| 2026 | GenPIE: A Time-Resolved Plenoptic Imager — Wang et al. | ACM TOG / SIGGRAPH 2026 | Reconstructs time-resolved plenoptic transport from sparse real transient measurements using generative 3D priors, differentiable transient path tracing, and spatially varying sensor-response calibration. It is tightly adjacent transient inverse rendering rather than a conventional around-corner method. |
| 2025 | Visible Occluders as Opportunistic Apertures for Wide Field of View Non-Line-of-Sight 3D Imaging — Czajkowski, Murray-Bruce | EUSIPCO 2025 | Uses ubiquitous doorway edges as a 180-degree computational aperture and improves passive depth recovery for hidden objects spanning substantial depth extents. |
| 2024 | Two-edge-resolved three-dimensional non-line-of-sight imaging with an ordinary camera — Czajkowski, Murray-Bruce | Nature Communications 2024 | TERI exploits two perpendicular doorway edges and one ordinary photograph of ceiling penumbrae for calibration-free, full-colour passive 3D reconstruction. |
| 2024 | Permutation Transient Attention Encoder for None-Line-of-Sight Imaging — Yue et al. | Optica Imaging Congress / COSI 2024 | Combines transient-attention blocks and Permute-MLP mixing for efficient, discriminative spatio-temporal feature extraction in learned active-NLOS reconstruction. |
| 2025 | High-resolution and real-time non-line-of-sight imaging based on spatial correlation — Zhang et al. | Optics and Lasers in Engineering 2025 | Develops SCBSF-NLOS, a scan-free non-confocal system with a 3D blur-kernel model and spatial-correlation resampling; reports about 2 cm lateral resolution and 5 fps dynamic reconstruction from a 16×16 detector. |
| 2024 | Real-time scan-free non-line-of-sight imaging — Zhang et al. | APL Photonics 2024 | Replaces relay-wall raster scanning with parallel SPAD-array acquisition and boundary migration, reaching 151 fps transient capture and 19 fps end-to-end imaging; super-resolution reduces the array requirement from 32×32 to 8×8. |
| 2025 | Forward and inverse diffraction in phasor fields - Garcia-Pueyo and Muñoz | Optics Express 2025 | Recasts phasor-field NLOS reconstruction as an inverse-diffraction problem, derives an explicit Inverse Phasor Fields operator, analyzes when the inverse is well posed, and proposes a diffraction-operator rank metric connected to the Rayleigh resolution criterion. |
| 2024 | Time-Gated Polarization for Active Non-Line-of-Sight Imaging - Pueyo-Ciutad et al. | SIGGRAPH Asia 2024 | Combines picosecond time-of-flight capture with polarization-aware transport and inversion, using directional information induced by the relay surface to recover hidden features inside the conventional missing cone with fewer relay-wall measurements. |
| 2026 | CUDA-accelerated Non-line-of-sight imaging with irregular relay surfaces - Sun et al. | Optics and Lasers in Engineering 2026 | Runs filtered back projection directly on non-planar relay geometry and arbitrary nonuniform source-detector samples, avoiding planarization or resampling; a CUDA implementation provides at least two orders of magnitude acceleration while retaining competitive measured-data reconstruction quality. |
| 2025 | Reprojection-Guided Non-Line-of-Sight Imaging Under Irregular Undersampling - Cui, Yue, Yang | IEEE JSTSP 2025 | Recovers dense denoised transients from sparse, irregular, or fragmented relay measurements; range-space reprojection supplies physics-grounded guidance that spatio-temporal modulation blocks inject adaptively, improving simulated-to-real and cross-relay generalization. |
| 2025 | Looking Around Flatland: End-to-End 2D Real-Time NLOS Imaging - Peña, Gutierrez, Marco | IEEE TCI 2025 | Reformulates transient light transport in self-contained 2D worlds and couples it to phasor-field camera models, enabling end-to-end real-time simulation, progressive reconstruction, and controlled analysis of filtering, reflectance, and geometry at up to five orders of magnitude lower cost than equivalent 3D experiments. |
| 2025 | Learning-based NLOS imaging with Kolmogorov-Arnold network-enhanced transformer - Liu et al. | Optics & Laser Technology 2025 | Combines a KAN-enhanced Transformer for nonlinear global transient features, convolutional attention for local detail, and EfficientSAM contour guidance, improving simulated and measured active-NLOS reconstructions. |
| 2025 | Feature enhanced non-line-of-sight imaging using graph model in latent space - Xu et al. | Optics & Laser Technology 2025 | Uses a group graph-Laplacian model in a compressed latent representation as a RED-style structural regularizer, balancing detail preservation, noise suppression, and efficiency for confocal and non-confocal transients. |
| 2025 | Non-line-of-sight imaging with adaptive artifact cancellation - Zhou et al. | Optics & Laser Technology 2025 | Introduces ground-truth-independent TOF-SSIM for parameter selection and adaptively cancels backprojection artifacts by modifying and reprojecting transient histograms in confocal and non-confocal settings. |
| 2025 | Adaptive Attention Based on Mixture Distribution for Zero-Shot Non-Line-of-Sight Imaging - Zhang, Liu, Duan | IEEE Signal Processing Letters 2025 | Models target and background residuals with a mixture distribution; dual-space adaptive weights act as zero-shot attention inside an alternating-minimization solver without paired training data. |
| 2025 | Passive non-line-of-sight pedestrian imaging based on light transport matrix decomposition - Chen et al. | Optics and Lasers in Engineering 2025 | Decomposes passive long-wave-infrared transport into illumination and reflection components and uses an illumination-oriented Transformer; IF-NLOS provides 33,000 images and experiments extend hidden-pedestrian imaging to 20 m. |
| 2025 | Non-line-of-sight imaging via scalable scattering mapping using TOF cameras - Fang et al. | Photonics Research 2025 | Uses a low-cost continuous-wave ToF camera and a learned scalable-scattering mapping to decouple object and relay-surface signatures, trained on 210,000 depth images and tested on laboratory and real walls. |
| 2026 | Edge-aware dual-spectral NLOS imaging architecture - Shi et al. | Proc. SPIE / NDTA 2025 (published 2026) | Fuses synchronized visible and long-wave-infrared measurements in an INT8 CPU pipeline, reporting 78.4 ms latency and sub-10 W operation on Jetson AGX Orin for deployable passive NLOS reconstruction. |
| 2025 | Ultrasound synthetic aperture non-line-of-sight imaging - Li et al. | Communications Physics 2025 | Transfers the core optical ToF/f-k reconstruction trajectory to coherent ultrasound: a phase-sensitive scanning emitter-receiver synthetic aperture reconstructs complex hidden 3D scenes at meter-scale range with approximately 1 cm lateral and depth resolution, using low-power eye-safe hardware. |
| 2026 | QSS-Net: A Quanta-State-Slot Network for Non-line-of-Sight Classification - Lin et al. | FLINS-ISKE / Springer LNCS 2026 | Formalizes a recognition-oriented NLOS direction with a Quanta-State-Slot network, shifting the target from complete hidden 3D recovery toward efficient semantic classification for time-sensitive perception. |
| 2026 | Multi-view clustering for non-line-of-sight imaging with neighborhood consistency reweighting - Lin et al. | Neurocomputing 2026 | Refines photon-corrupted neighborhood graphs across multiple NLOS reconstructions using shared-neighbor consistency, Huber-smoothed dual consistency gates, and adaptive view weighting, with closed-form alternating updates and convergence guarantees. |
| 2025 | Adaptive motion enhancement for passive non-line-of-sight action recognition - Sun et al. | Neurocomputing 2025 | Introduces AME-Net for recognizing hidden human actions from ordinary RGB videos of a visible relay wall and releases NLOS-Action, the first passive NLOS action-recognition dataset, with 2,000 synthetic and more than 500 real sequences. |
| 2025 | Non-line-of-sight multi-person pose sensing - Hou et al. | Optics Express 33(20), 41937–41950 (2025) | Extends transient NLOS semantics from single-person pose to adaptive multi-person 3D mesh recovery. AMPE-NLOS combines LCT-propagated coarse features, 3D U-Net refinement, body-center heatmaps, and SMPL parameter maps; simulated and measured confocal SPAD experiments cover varying hidden-person counts. |
| 2026 | Non-line-of-sight human pose estimation - Xiao et al. | Optics and Lasers in Engineering 2026 | Directly estimates semantic body joints from transient-derived intensity and depth features, using a physics-based smartphone-video synthesis pipeline for training and demonstrating robust measured-data pose recovery at 1.75 m and SNR 0.13. |
| 2026 | Frequency-domain multi-regularization-experts fusion for robust non-line-of-sight imaging - Zhang et al. | Pattern Recognition 2026 | Introduces a frequency-domain mixture of regularization experts whose gating weights are derived from optimization dual variables, combining direct-method efficiency with improved robustness and real-data albedo reconstruction. |
| 2026 | Canny operator-based artifact identification and suppression for non-line-of-sight imaging - Chen et al. | Optics and Laser Technology 2026 | Combines Canny/morphological artifact segmentation, reference-free sharpness and residual metrics, and genetic optimization of a transient blur kernel to suppress backprojection artifacts in simulated and measured confocal and non-confocal data. |
| 2026 | Nonconfocal non-line-of-sight imaging with specular-flight-path regularization for complex multi-orientation objects - Shi et al. | Photonics Research 2026 | Uses local matrix analysis to identify low-redundancy specular-flight-path regions and regularizes non-confocal inversion around those paths, improving reconstruction of tilted, multi-orientation hidden objects. |
| 2026 | Super-resolution non-line-of-sight imaging with laser pulses multiplexing - Miao et al. | Optics and Lasers in Engineering 2026 | Uses deterministic cross-pulse sub-bin modulation and non-negative least squares to recover 64 ps transients from 704 ps photon-timing hardware, retaining benefits under severe under-scanning and extending to non-confocal arrays. |
| 2026 | Finite-Aperture Limits for Yaw Estimation in Confocal Non-Line-of-Sight Imaging - Romanelli et al. | Journal of Imaging 2026 | Derives a finite-relay-wall switch-line criterion and Fisher-information bounds for yaw observability in confocal transient NLOS. Simulated and measured f-k/backprojection results show how angular information degrades as informative transient features are clipped, providing quantitative guidance for relay-aperture placement and orientation recovery. |
| 2026 | Machine Learning-Based Human Detection Using Active Non-Line-of-Sight Laser Sensing - Celebi, Turkoglu | Sensors 2026 | Uses real SPAD-TCSPC transient measurements from controlled rubble-like scenes to compare CNN, GRU, and random-forest classifiers for hidden-human presence across poses, orientations, and object layouts. All models reach full sensitivity, while random forest provides the strongest specificity and weighted F1 under limited photon counts. |
| 2026 | Adaptive Spiral Scanning for Confocal Non-Line-of-Sight Imaging - Oyama et al. | IEEE OJSP 2026 | Dynamically shifts an Archimedean spiral toward relay-wall regions with strong measured return intensity and applies Voronoi density compensation to correct the resulting nonuniform sampling, extending transient-sinogram acquisition from fixed circular trajectories to adaptive measurement allocation. |
| 2026 | Submillimeter non-line-of-sight ranging and imaging via cost-effective FMCW interferometry - Liang et al. | Photonics Research 2026 | Uses a dual-path FMCW interferometer with fixed-delay fiber calibration and dynamic temporal phase subdivision to compensate sweep nonlinearity, demonstrating 450-um ranging resolution, millimeter-scale 3D NLOS imaging, and operation under more than 8 klux ambient illumination. |
| 2026 | Breaking the speed-resolution trade-off in 3.3-km non-line-of-sight imaging using scanning-free laser reflective tomography - Wang et al. | Opto-Electronic Science 2026 | Treats the diffuse relay wall as a natural beam expander and combines single-point third-bounce detection with multi-angle laser reflective tomography. The scanning-free system reports roughly 91x faster acquisition, 3.3-km operation, and better-than-3-cm resolution in about three minutes. |
| 2026 | Non-confocal non-line-of-sight imaging using frequency-domain phase compensation with the reference function - Yu et al. | Optics Express 2026 | Transfers a single-input/multiple-output millimeter-wave frequency-domain imaging formulation to optical non-confocal NLOS and uses reference-function phase compensation plus FFT reconstruction to reduce artifacts, shape distortion, and computation. |
| 2026 | Structure-guided adaptive total variation for parameter-free passive non-line-of-sight imaging - Zhang et al. | Optics Express 2026 | Derives spatially varying total-variation weights from a preliminary hidden-scene estimate, automatically balancing edge preservation and noise suppression for conventional-camera passive NLOS without manual regularization tuning. |
| 2026 | Non-Line-of-Sight Imaging via Sparse Bayesian Learning Deconvolution - Tian et al. | Photonics 2026 | Introduces a geometry-agnostic sparse Bayesian transient-restoration front end before LCT or f-k migration, suppressing photon background and instrument-response blur while preserving physically meaningful multipath returns. |
| 2026 | Non-Line-of-Sight Single-Pixel Imaging Using Polarization Speckle Modulation - Zhou et al. | Physical Review Letters 2026 | Modulates polarization to generate diverse rough-wall speckle illuminations and recovers hidden scenes with a single-pixel detector, enabling scanning-free millimeter-resolution keyhole imaging with noninvasive memory-effect calibration. |
| 2026 | High-resolution Fourier single-pixel non-line-of-sight imaging employing diffusion model - Fu et al. | Optics and Lasers in Engineering 2026 | Uses measured low-frequency Fourier coefficients as a consistency constraint inside diffusion sampling, restoring high-frequency hidden-scene detail at sampling rates as low as 3 percent under strong multilayer scattering. |
| 2026 | Multiple-object passive non-line-of-sight imaging - Chen et al. | Frontiers of Computer Science 2026 | Introduces a nested U-Net multi-attention model with channel, spatial, and self-attention plus multi-branch supervision to reconstruct passive infrared scenes containing one to three simultaneously hidden objects. |
| 2025 | Passive non-line-of-sight imaging at 10 meters — Zhou et al. | Optics Letters 2025 | Combines pattern-based calibration with low-rank matrix decomposition to enhance weak wall-encoded signals and separate ambient background. The ordinary-camera system demonstrates passive NLOS at a 10 m wall-to-camera distance and 4.5 m wall-to-object distance, with centimeter-scale resolution at an SBR as low as −13 dB. |
| 2026 | Neural Networks Meet Light Transport Physics for Passive Non-Line-of-Sight Imaging Enhancement — Liang et al. | IEEE TCI 2026 | Introduces HPDI, a dual-path hybrid of a physics-informed coarse-to-fine pathway and a data-driven implicit reconstruction pathway, followed by adaptive fusion. It improves fidelity, generalization, and data efficiency across passive scenes with and without occluders while retaining an interpretable light-transport branch. |
| 2026 | Super-field-of-view non-line-of-sight imaging via spatial encoding of a translated point spread function — Li et al. | Photonics Research 2026 | Translates the transient point-spread function and spatially encodes the corresponding convolution to reconstruct targets outside the physical scan region without additional capture. Simulations cover 9× the detection area, while experiments reconstruct beyond twice the single-axis detection range, equivalent to about 19.6× the original area after translated reconstruction and stitching. |
| 2026 | Stereo non-line-of-sight imaging — Luesia-Lahoz, Cartiel, Muñoz | The Visual Computer 2026 | Introduces a two-relay-wall stereo configuration in which phasor fields treat both walls as generalized virtual apertures. Combining same-wall and cross-wall illumination/detection paths reduces missing-cone ambiguity and provides surface-orientation cues that are unavailable from a single relay wall. |
| 2026 | Compact non-line-of-sight imager at long range — Zeng et al. | Optics Express 2026 | Integrates optimized collection optics, adaptive scan-synchronized gating, a gated InGaAs/InP SPAD, and fast electronics into a compact outdoor prototype. It demonstrates daylight kilometer-scale NLOS, approximately 4 cm spatial resolution, and 2 fps imaging of simple moving targets. |
| 2026 | A model decomposition method for the real-time non-line-of-sight imaging — Yang et al. | iScience 2026 | Reformulates sparse transient NLOS reconstruction as a spectrally filtered non-negative LASSO problem and decomposes the optimization into GPU-parallelizable subproblems. The reported system reconstructs 128×128 hidden images from 64 measurements in 4.6 s. |
| 2026 | Passive non-line-of-sight imaging with diffuse-aware attention-enhanced encoding — Wang et al. | Optics Express 2026 | Introduces diffuse-aware, attention-enhanced feature encoding for ordinary-camera passive NLOS reconstruction, explicitly targeting information loss caused by diffuse relay-wall transport rather than treating the wall projection as an unconstrained image-to-image mapping. |
| 2026 | Learned light-cone transform for fast and accurate non-line-of-sight imaging — Gao et al. | Physical Review Applied 2026 | Citation-traced directly from the original LCT: replaces its fixed low-pass Wiener stage with high-frequency-enhanced learnable filtering, spatiotemporal feature extraction, and volume projection; reports a 5 dB PSNR gain over LCT and cross-system measured-data generalization. |
| 2026 | Transient video interpolation for dynamic non-line-of-sight imaging — Sun et al. | Optics Express 2026 | Introduces TransVID, the first diffusion-based transient-video interpolation framework for dynamic NLOS; jointly performs spatial and temporal upsampling, recovering 128×128 transient video at 16 fps from 16×16 measurements at 4 fps before hidden-scene reconstruction. |
| 2025 | Multi-surface sub-resolution non-line-of-sight imaging via transient waveform deposition — Wei et al. | Optics Express 2025 | Decomposes overlapping multi-surface transient returns as Gaussian mixtures using L1–L2 deconvolution and constrained modified Levenberg–Marquardt fitting, then reconstructs each deposited waveform with LCT; improves reported axial/lateral discrimination from 5.25/9.74 cm to 1 cm. |
| 2026 | MDUNet: Multimodal Decoding UNet for Passive Occluder-Aided Non-line-of-sight 3D Imaging — Raji, Murray-Bruce | WACV 2026 | Jointly reconstructs a 3D hidden-occluder SDF/mesh and a 2D radiosity image of non-occluding content from one ordinary soft-shadow photograph. Shared latent features couple the two decoder paths, delivering over 100× faster inference than Soft Shadow Diffusion and over 1000× faster inference than iterative optimization while transferring from simulation to real measurements. |
| 2026 | Thermal Non-Line-of-Sight Imaging through Rough Surfaces — Ye et al. | ACM TOG 2026 | Introduces NLOSFormer for passive thermal imaging through rough relay surfaces. An explicitly estimated convolution kernel constrains an end-to-end reconstruction network; ThermalNLOS supplies training and evaluation data, and experiments demonstrate rough-wall generalization, relative depth estimation, and dynamic reconstruction at 4 fps. |
| 2026 | Non-line-of-sight imaging via physics-informed cascade learning — Zhao et al. | JOSA A 2026 | Cascades a lightweight SPAD-specific noise-separation network with a reconstruction network containing a differentiable NLOS forward model. The self-supervised physics-informed design removes the need for large paired datasets and improves robustness under mixed detector noise and low-SNR measurements. |
| 2025 | Adaptive windowing for photon-efficient non-line-of-sight imaging under high ambient light — Miao et al. | Optics Express 2025 | Models signal and background photon arrivals, clusters correlated pixels to estimate matched-filter windows, and applies per-pixel short temporal gates with total-variation reconstruction. AW-NLOS remains effective at SBR 2.12 and 0.02 detected signal photons per pixel, improving SSIM by approximately 0.5 at 42 mW laser power. |
| 2026 | All-day non-line-of-sight imaging based on Si-SPAD and phase-congruency-based structured ε-regularization — Yin et al. | Optics and Lasers in Engineering 2026 | Co-designs a detector-aware Si-SPAD system and phase-congruency structured regularization for extreme ambient light. The system reports an 18× SBR gain over InGaAs-SPAD capture and demonstrates 200 m NLOS imaging under 94,314 lx illumination with 4 cm lateral and 1 cm axial resolution. |
| 2017 | Fast Back-Projection for Non-Line of Sight Reconstruction — Arellano, Gutierrez, Jarabo | Optics Express 2017 | Recasts classical transient back-projection as GPU voxelization of the ellipsoidal space--time manifolds induced by three-bounce measurements, reducing redundant voxel-wise evaluation and delivering up to three orders of magnitude faster reconstruction with negligible quality loss. |
| 2024 | Toward Dynamic Non-Line-of-Sight Imaging with Mamba Enforced Temporal Consistency — Li et al. | NeurIPS 2024 | Introduces ST-Mamba, a spatial--temporal state-space architecture for dynamic active NLOS reconstruction; interleaved Mamba blocks model long transient/video dependencies efficiently, while a phasor-field wave-domain loss enforces measurement physics and improves temporal consistency under noise. |
| 2023 | Enhancing Non-Line-of-Sight Imaging via Learnable Inverse Kernel and Attention Mechanisms — Yu et al. | ICCV 2023 | Introduces a system-aware learned inverse kernel tailored to transient NLOS point-spread functions, together with attention mechanisms that refine hidden intensity and depth reconstruction; it forms a direct learned-physics bridge between shared feature embeddings and later transformer/generalizable-prior models. |
| 2020 | Learned Feature Embeddings for Non-Line-of-Sight Imaging and Recognition — Chen et al. | ACM TOG / SIGGRAPH Asia 2020 | Introduces a physics-structured learned embedding that maps transient measurements into a shared hidden-scene representation for high-resolution reconstruction, classification, and 2.5D detection; differentiable propagation, visibility, rendering, and depth modules enable synthetic-to-real generalization on measured NLOS scenes. |
| 2023 | Deep Non-line-of-sight Imaging from Under-scanning Measurements — Li et al. | NeurIPS 2023 | Introduces the first deep-learning pipeline designed specifically for under-scanned active confocal transients: a transient-recovery network restores a dense measurement grid and a physics-aware volume-reconstruction network recovers hidden 3D geometry; it remains effective at 8×8 scans and reports 50× faster inference than the iterative baseline. |
| 2019 | Beyond Volumetric Albedo -- A Surface Optimization Framework for Non-Line-of-Sight Imaging — Tsai, Sankaranarayanan, Gkioulekas | CVPR 2019 | Moves active transient NLOS beyond voxelized albedo by directly optimizing a hidden surface and reflectance. Its differentiable radiometric renderer supplies geometry- and reflectance-aware gradients, while stochastic optimization and geometry processing recover substantially finer surfaces than prior volumetric methods. |
| 2020 | Non-Line-of-Sight Reconstruction Using Efficient Transient Rendering — Iseringhausen, Hullin | ACM TOG 2020 | Establishes a physically grounded analysis-by-synthesis route for active transient NLOS: a custom millisecond three-bounce renderer, implicit level-set surface geometry, BRDF and visibility terms, and global optimization recover detailed hidden surfaces under noise and non-diffuse reflectance, forming a direct precursor to differentiable, neural-field, and Gaussian transient rendering. |
| 2021 | Towards Non-Line-of-Sight Photography — Peng et al. | arXiv 2021 | Reframes active transient NLOS as direct image synthesis: a data-driven model maps measured transients to a high-resolution 2D photograph from the relay-wall viewpoint, bypassing an explicit intermediate 3D reconstruction and emphasizing hidden appearance and texture rather than geometry alone. |
| 2023 | Fast Non-Line-of-Sight Imaging with Non-Planar Relay Surfaces — Gu et al. | IEEE ICCP 2023 | Extends Rayleigh--Sommerfeld / phasor-field reconstruction to non-planar relay surfaces through two-stage wave propagation with a planar proxy, preserving fast wave-based inversion when the visible relay geometry is curved or irregular. |
| 2023 | Non-Line-of-Sight Imaging With Signal Superresolution Network — Wang et al. | CVPR 2023 | Learns a plug-and-play transient superresolution operator that recovers dense measurement grids from sparse confocal or non-confocal scans, cutting acquisition time by 16× while retaining comparable reconstruction quality. |
| 2023 | Virtual Mirrors: Non-Line-of-Sight Imaging Beyond the Third Bounce — Royo et al. | ACM TOG / SIGGRAPH 2023 | Treats planar diffuse hidden surfaces as computational phasor-domain ‘virtual mirrors’, builds secondary virtual apertures from higher-order bounces, and reconstructs limited-visibility geometry and objects hidden behind a second corner without requiring physical specular reflectors. |
| 2023 | Fast Differentiable Transient Rendering for Non-Line-of-Sight Reconstruction — Plack et al. | WACV 2023 | Introduces a fast differentiable transient renderer that reduces analysis-by-synthesis NLOS inverse-rendering runtimes from hours to minutes on consumer hardware, improves optimization stability, and enables self-supervised transient reconstruction. |
| 2023 | Omni-Line-of-Sight Imaging for Holistic Shape Reconstruction — Huang et al. | arXiv 2023 | Unifies straight-ray LOS and spherical-wavefront NLOS transient rendering in a neural level-set model, using nearby diffuse walls as virtual mirrors to recover near-360° object geometry from one fixed scan position; validated with a SPAD/laser prototype. |
| 2018 | Revealing hidden scenes by photon-efficient occlusion-based opportunistic active imaging — Xu et al. | Optics Express 2018 | Converts the occlusion-coding theory into a photon-efficient room-scale system: a binomial single-photon forward model reconstructs hidden reflectivity from non-time-resolved three-bounce counts and needs about 16× fewer detected photons than the earlier Gaussian model. |
| 2019 | Enhancing Passive Non-Line-of-Sight Imaging Using Polarization Cues — Tanaka, Mukaigawa, Kadambi | arXiv 2019 | Adds a camera-mounted polarizer to passive NLOS capture: polarization-axis rotation makes oblique indirect paths more discriminative, improves the conditioning of the light-transport inverse problem, and enhances hidden-image recovery both with and without occluders. |
| 2020 | Seeing Around Corners with Edge-Resolved Transient Imaging — Rapp et al. | Nature Communications 2020 | Combines a vertical edge occluder with pulsed SPAD transients: differencing adjacent photon-arrival histograms isolates angular wedges and enables 2.5D room-scale reconstruction over a 180° field of view from only 45 illumination positions. |
| 2026 | X-band Radar Non-Line-of-Sight Imaging — Du et al. | CVPR 2026 | Introduces a learned, geometry-aware X-band radar NLOS reconstruction system; converts normally diffuse optical transport into predominantly specular long-wavelength transport and experimentally reconstructs hidden objects at ranges up to 40 m. |
| 2026 | Neural illumination fields: High-fidelity and ambient-robust stereo reconstruction for two-bounce non-line-of-sight imaging — Zhang et al. | Optics and Lasers in Engineering 2026 | Uses a self-supervised continuous neural illumination field and differentiable intensity rendering for two-bounce NLOS reconstruction without binary shadow segmentation, reaching centimeter-scale detail in large hidden spaces under low-contrast and ambient-light interference. |
| 2026 | CUDA-accelerated non-line-of-sight imaging with irregular relay surfaces — Sun et al. | Optics and Lasers in Engineering 2026 | GPU-accelerated filtered backprojection for measurements acquired on irregular, non-planar relay surfaces, reducing the practical cost of geometry-aware active NLOS reconstruction. |
| 2025 | Fast Non-Line-of-Sight Transient Data Simulation and an Open Benchmark Dataset — Shi et al. | Optics Express 2025 | Provides a configurable intensity-transport transient simulator from depth and optional albedo maps, models detector jitter and Poisson noise, releases seven ShapeNet-category datasets, and benchmarks LCT, phasor-field, f-k, and backprojection baselines. |
| 2026 | Geometry-Constrained Non-Line-of-Sight Imaging — Liu et al. | IEEE TVCG 2026 | Jointly reconstructs volumetric albedo, depth, and hidden-surface geometry; a Frobenius-norm shape-operator prior regularizes surface-normal variation, improving sparse/short-exposure reconstruction while reporting roughly 30× lower optimization time than an existing surface-reconstruction baseline. |
| 2025 | TransDiff: Unsupervised Non-Line-of-Sight Imaging with Aperture-Limited Relay Surfaces — Cui et al. | IEEE TIP 2025 | Uses latent diffusion and unsupervised measurement consistency to synthesize/recover dense transient information from aperture-limited relay measurements. |
| 2025 | Under-scanning non-line-of-sight imaging based on convolution approximation and optimization — Miao et al. | APL Photonics 2025 | DO-NLOS combines a convolution approximation with optimization to reconstruct hidden scenes from strongly under-scanned transient measurements. |
| 2025 | Physics to the Rescue: Deep Non-line-of-sight Reconstruction for High-speed Imaging — Mu et al. | IEEE TPAMI 2025 | Embeds wave-propagation and volume-rendering priors in a neural model for robust high-speed NLOS reconstruction from approximate real-time capture models. |
| 2025 | Sub-pixel resolving modulation for non-line-of-sight imaging — Zhang et al. | Optics Express 2025 | Introduces sub-pixel modulation to improve spatial resolving power beyond the native relay-wall sampling grid. |
| 2024 | Miniaturized time-correlated single-photon counting module for time-of-flight non-line-of-sight imaging applications — Wu et al. | Review of Scientific Instruments 2024 | Miniaturizes the timing backend of single-photon ToF NLOS: a four-channel TCSPC module provides a 10 ps bin and 27.4 ps minimum RMS timing resolution, supporting 6.3 cm lateral and 2.3 cm depth resolution at a 5 m sensor distance. |
| 2024 | NIGHT -- Non-Line-of-Sight Imaging from Indirect Time of Flight Data — Caligiuri et al. | ECCV 2024 Workshops, Part XIX | Uses an off-the-shelf indirect-ToF sensor and a learned virtual-mirror representation to infer hidden-scene depth without custom direct-ToF hardware; the accompanying synthetic dataset supports the first consumer-iToF NLOS benchmark. |
| 2024 | Virtual Scanning: Unsupervised Non-line-of-sight Imaging from Irregularly Undersampled Transients — Cui et al. | NeurIPS 2024 | Learns from irregularly undersampled transients without paired supervision, combining virtual scanning, measurement consistency, and a physics-guided SURE denoiser. |
| 2024 | Real-time non-line-of-sight computational imaging using spectrum filtering and motion compensation — Ye et al. | Nature Computational Science 2024 | Couples spectrum-domain filtering with motion compensation/interleaved scanning to deliver room-scale real-time NLOS video. |
| 2024 | Plug-and-play algorithms for dynamic non-line-of-sight imaging — Ye et al. | ACM TOG 2024 | A flexible dynamic-NLOS inverse framework that combines wave-physics filtering with learned denoiser priors for temporally consistent hidden-scene video. |
| 2024 | Domain Reduction Strategy for Non Line of Sight Imaging — Shim et al. | ECCV 2024 | Periodically prunes empty hidden-space regions during inverse rendering, accelerating geometry/albedo recovery under sparse and general acquisition setups. |
| 2025 | Non-Line-of-Sight Imaging Using Raster Scanning at NIR Wavelength — Roueinfar, Salmanian | IEEE ICEE 2025 | Uses a 500 mW, 808 nm laser on a pan-tilt raster scanner and an NIR camera to recover three hidden targets through steady-state three-bounce transport. It is a low-cost acquisition baseline rather than a new inverse-model frontier; the repository uses the final IEEE conference record instead of the later arXiv upload. |
| 2026 | Non-line-of-sight imaging with arbitrary relay surface geometries via 3D Gaussian Transient Rendering — Wang et al. | SIGGRAPH 2026 | Uses 3D Gaussian primitives and differentiable transient rendering; targets spatially limited, non-planar, arbitrary relay surfaces. |
| 2026 | Imaging Hidden Objects with Consumer LiDAR via Motion Induced Sampling — Somasundaram et al. | Nature 2026 | Demonstrates plug-and-play NLOS on smartphone-grade consumer LiDAR through motion-induced aperture sampling and multi-frame fusion, supporting hidden 3D reconstruction, single/multi-object tracking, and camera localization with off-the-shelf hardware. |
| 2026 | Seeing through boxes: Non-Line-of-Sight 3D Reconstruction from Radar Signals — Lu et al. | CVPR 2026 | GeRaF 2.0 combines visible exterior geometry with NLOS RF propagation in a unified neural field, stabilizing signed-distance optimization and producing physically consistent visible and hidden surfaces. |
| 2024 | Feasibility of Non-Line-of-Sight Integrated Sensing and Communication at mmWave — Tosi et al. | IEEE SPAWC 2024, 331–335 | Demonstrates fully NLOS target detection with a 27.4-GHz 5G/mmWave ISAC proof-of-concept in a factory-like environment; CSI processing suppresses TDD-induced spectral replicas and establishes the experimental precursor to the 2026 ICT intrusion-detection and tracking system. |
| 2026 | Reliable Non-Line-of-Sight Intrusion Detection with Integrated Sensing and Communications Hardware — Tosi et al. | International Conference on Telecommunications (ICT 2026), 25–30 | Uses 5G/mmWave ISAC hardware and large-surface reflections for fully occluded industrial intrusion sensing; range–Doppler processing and PHD-based tracking improve target persistence and false-alarm robustness beyond the earlier feasibility study. |
| 2026 | Radar Cross Section Characterization of Quantized Reconfigurable Intelligent Surfaces — Yasmeen et al. | arXiv 2026 | Complements RIS around-corner radar sensing by characterizing one-bit/quantized RIS RCS and experimentally recovering micro-Doppler signatures in non-specular or shadowed regions. |
| 2026 | DENALI: A Dataset Enabling Non-Line-of-Sight Spatial Reasoning with Low-Cost LiDARs — Behari et al. | CVPR 2026 | Provides a large-scale low-cost LiDAR space-time-histogram dataset and benchmark for data-driven NLOS spatial reasoning, lowering the barrier to learning with commodity depth sensors. |
| 2026 | NLOS-Aided Joint OTA Synchronization and Off-Grid Imaging for Distributed MIMO Systems — Tong et al. | arXiv 2026 | Jointly refines over-the-air synchronization and sparse off-grid environment imaging in distributed MIMO/ISAC by exploiting reconstructed NLOS components. |
| 2026 | A comprehensive study of time-of-flight non-line-of-sight imaging — Marco et al. | arXiv 2026 | Benchmark-style comparative study unifying ToF NLOS forward/inverse models under common hardware constraints. |
| 2026 | Exploiting Double-Bounce Paths in Snapshot Radio SLAM: Bounds, Algorithms and Experiments — Zhang et al. | arXiv 2026 | Shows that higher-order NLoS radio paths can improve snapshot mmWave SLAM and reveal landmarks hidden to single-bounce models. |
| 2026 | Cramer-Rao Bounds for Target Parameter Estimation in a Bi-Static IRS-Assisted Radar Configuration — Sanjeeva Reddy S, Vinod Veera Reddy | arXiv 2026 | Characterizes angle-estimation limits for a passive-IRS bistatic NLOS radar geometry, clarifying what reconfigurable RF relay surfaces can theoretically recover. |
| 2024 | Around-the-corner Radar Sensing Using Reconfigurable Intelligent Surface — Yasmeen et al. | IEEE MAPCON 2024 | RIS-assisted around-corner radar sensing; steers energy into NLOS regions and recovers human micro-Doppler signatures. |
| 2025 | Radar Sensing using Dual-Beam Reconfigurable Intelligent Surface — Yasmeen et al. | IEEE RadarConf25 2025 | Benchmarks a practical 1-bit dual-beam RIS against a metal plate and an ideal single-beam RIS in simulations and measurements, widening NLOS radar coverage and supporting simultaneous multi-direction sensing. |
| 2025 | Non-Line-of-Sight Human Vital-Sign Sensing Aided by Reconfigurable Intelligent Surfaces — Li et al. | Acta Electronica Sinica 53(1), 1–13 (2025) | Programmable-RIS NLOS physiological sensing: vision guides metasurface beam control toward the hidden chest region, followed by respiration/heartbeat estimation. |
| 2025 | Liquid Crystal RIS Integrated with SIL Radar for NLOS Vital Sign Monitoring — Tripathy et al. | IEEE IMBioC 2025 | Liquid-crystal RIS plus self-injection-locked radar for experimentally validated hidden-region vital-sign monitoring. |
| 2026 | Beyond λ/2: Can Arbitrary EMVS Arrays Achieve Unambiguous NLOS Localization? — Chen et al. | arXiv 2026 | RIS-aided NLOS bistatic MIMO radar localization with EMVS polarization cues and phase-disambiguating array geometry. |
| 2026 | RIS-aided Radar Detection Architectures with Application to Low-RCS Targets — Colone et al. | arXiv 2026 | Designs RIS-aided monostatic/bistatic radar detection architectures that redirect low-RCS target energy back toward the radar. |
| 2026 | Backscatter Assisted Indoor NLOS Positioning — Ruttik et al. | arXiv 2026 | Uses passive backscatter devices as virtual anchors for RF NLOS indoor positioning in corridors. |
| 2026 | Ambient IoT Backscatter Devices as Passive Anchors for NLOS Cellular Positioning: Fundamental Limits — Yiğitler et al. | arXiv:2607.03459 (2026) | Derives closed-form equivalent Fisher information matrices for calibrated, partially calibrated, and fully uncalibrated Ambient-IoT backscatter anchors in uplink NLOS positioning. It quantifies which carrier-phase and delay information survives unknown device phases/gains and shows that joint single-snapshot UE–scatterer identifiability requires at least two passive anchors in 2D or three in 3D with sufficient angular diversity. |
| 2024 | Soft Shadow Diffusion (SSD): Physics-Inspired Learning for 3D Computational Periscopy — Raji, Murray-Bruce | ECCV 2024 | Decomposes a hidden scene into light-occluding 3D geometry and non-occluding radiosity, yielding a separable nonlinear least-squares inverse; a physics-inspired diffusion network trained only in simulation transfers to real soft-shadow measurements. |
| 2024 | Towards 3D Computational Persicopy with an Ordinary Camera: A Separable Non-Linear Least Squares Formulation — Raji, Murray-Bruce | IEEE ICASSP 2024 | Introduces the separable nonlinear least-squares formulation that jointly estimates coarse 3D light-occluding geometry and a 2D non-occluding reflectance plane from one ordinary soft-shadow photograph, providing the optimization foundation later used by SSD. |
| 2024 | Two-Edge-Resolved 3D Non-Line-of-Sight Imaging: A Fisher Information Equalized Discretization — Czajkowski, Murray-Bruce | IEEE ICASSP 2024 | Uses Fisher-information-equalized range discretization for two-edge-resolved ordinary-camera NLOS, allocating depth samples according to measurement sensitivity and refining the numerical conditioning of the TERI reconstruction pipeline. |
| 2025 | Enabling NLOS Imaging Capabilities at the Initial Access of 6G Base Stations — Tornielli Bellini et al. | arXiv 2025 | Integrates coherent reflector-assisted NLOS imaging into the standard initial-access beam sweep of a next-generation base station; jointly designs a static modular reflector and imaging-specific codebook, derives near-field resolution/effective-aperture and moving-target trade-offs, and validates the system experimentally. |
| 2025 | MITO: Enabling Non-Line-of-Sight Perception using Millimeter-waves through Real-World Datasets and Simulation Tools — Dodds et al. | arXiv 2025 | Dataset and simulation tooling for mmWave NLOS perception with paired LOS/NLOS captures and RGB-D/mask supervision. |
| 2025 | See and Beam: Leveraging LiDAR Sensing and Specular Surfaces for Indoor mmWave Connectivity — Bandari et al. | arXiv 2025 | Uses LiDAR-visible specular surfaces to infer NLOS mmWave reflection paths and localize users for beam steering. |
| 2025 | 3D Reconstruction from Transient Measurements with Time-Resolved Transformer — Li et al. | arXiv 2025 | TRT/TRT-NLOS spatio-temporal transformer for photon-efficient LOS/NLOS transient 3D reconstruction with code and datasets. |
| 2025 | TransiT: Transient Transformer for Non-line-of-sight Videography — Li et al. | ICCV 2025, 27542–27551 | Transient transformer for high-speed NLOS videography; reconstructs 10 fps videos from sparse
Key breakthroughs that shaped the NLOS Imaging field:
2008 ── Raskar & Davis: 5D Time-Light Transport — theoretical foundation
│
2012 ── Velten et al.: first experimental 3D NLOS reconstruction [Nature Comm.]
│
2014 ── Heide et al.: Diffuse Mirrors — ToF camera + optimization [SIGGRAPH]
│
2015 ── Buttafava et al.: SPAD-based NLOS — more accessible hardware
│
2017 ── Arellano et al.: fast GPU back-projection — ellipsoidal space-time manifold voxelization brings classic transient reconstruction from hours toward seconds [Optics Express]
│
2018 ── O'Toole et al.: confocal NLOS + LCT — real-time O(N³logN) [Nature]
│ Liu et al.: phasor field — NLOS as virtual LOS wave propagation
│
2019 ── Tsai et al.: Beyond Volumetric Albedo — direct hidden-surface and reflectance optimization [CVPR]
│ Seidel et al.: a single wall edge becomes a natural aperture, enabling one-photograph angular hidden-scene estimation [IEEE ICCP]
│ Musarra et al.: time-resolved single-pixel capture demonstrates sub-second full-colour 3D NLOS imaging [OSA Imaging Systems and Applications]
│ Dove and Shapiro: paraxial phasor-field theory — TFSWD propagation formalizes occlusions and specular transport beyond the P-field alone [Optics Express]
│ Reza et al.: experimental phasor-field waves — interference and independent optical/P-field focusing validate the virtual-wave picture [Optics Express]
│
2020 ── Iseringhausen & Hullin: physically based transient analysis-by-synthesis — surface-, BRDF-, and visibility-aware NLOS inverse rendering [ACM TOG]
│ Sasaki and Leger formulate wall-scattered plenoptic inversion with a BRDF-aware Fredholm model and explicit limits on retrievable hidden-scene information [JOSA A]
│ Saunders et al.: multi-depth computational periscopy recovers two hidden images together with their wall-relative depths from one ordinary photograph [IEEE ICASSP]
│ Takatori: stereo virtual-image geometry establishes reflected-obstacle NLOS localization for intelligent vehicles [IEEE IV]
│ Dove and Shapiro: physical and nonparaxial phasor optics — ordinary lenses manipulate P-fields and Rayleigh--Sommerfeld theory extends them beyond Fresnel geometry [Optics Express]
│ Chen et al.: learned NLOS feature embeddings — a shared physics-aware representation for reconstruction, detection, and recognition [SIGGRAPH Asia / TOG]
│
2025 ── mmNorm: mmWave surface-normal estimation advances hidden RF sensing from localization to 3D object geometry [MobiSys]
│ Jeon et al.: ray-traced 2D mmWave point clouds localize multiple hidden pedestrians at measured outdoor T-junctions [IEEE IV]
│ Zhai et al.: second-order edge diffraction plus block sparse Bayesian learning localizes hidden acoustic sources without direct/first-order paths [Applied Acoustics]
│ BiScalar-AA and TSAN: attention-based radar pseudo-image models extend NLOS toward learned object detection and tracking [IEEE SWC / Computer Engineering]
│ Zheng et al.: sectionalized-ellipsoid interpolation converts SPAD-array non-confocal data into semi-confocal histograms [Optics and Lasers in Engineering]
│ Yu et al.: spherical-slice transform gives a fast high-resolution non-confocal frequency-domain inverse [Optics Letters]
2025 ── Chen et al.: hierarchical-NeRF implicit ray carving makes two-bounce shadow reconstruction more efficient [Optics Express]
│ Christnacher et al.: acquisition-parameter sensitivity moves real-time NLOS optimization upstream, quantifying SPAD pixel calibration, exposure, frame count, field of view, and focus before downstream f-k/back-projection/phasor-field inversion [Proc. SPIE]
│ A unified geometric-/physical-optics light-field analysis connects roughness, wavelength, diffraction, and focusing across passive LWIR and THz NLOS [Optics Express]
│ Zhang et al.: DCPNet advances two-bounce NLOS from reconstruction to real-time dual-view tracking [Optics Express]
│ 2025 ── Roueinfar & Salmanian: low-cost 808 nm steady-state NIR raster scanning [IEEE ICEE]
│ Tian et al.: deconvolution and virtual modulated range migration recover 50× super-resolution histograms from nanosecond-resolution ToF data [Optics Letters]
│ Guo et al.: vectorial digitelligent optics jointly optimizes phase and polarization to refocus through a rough relay wall and attain near-diffraction-limited 0.40-mm NLOS imaging [Engineering]
│ Oyama and Takatori: painted-surface reflections support semantic NLOS obstacle detection after reflection-aware appearance normalization [IEEE ITSC]
│ Yuan et al.: RIS-enabled covariance-domain gridless DoA uses atomic-norm recovery and ADMM for simulated multi-target NLOS angular sensing [Signal Processing]
│ Li et al.: programmable RIS beam control enables hidden-region respiration and heartbeat sensing [Acta Electronica Sinica]
│ Tripathy et al.: liquid-crystal RIS + SIL radar experimentally monitors NLOS vital signs [IEEE IMBioC]
│ Park et al.: camera-derived T-junction geometry conditions mmWave multipath for hidden-pedestrian localization [IEEE/RSJ IROS]
│ Jeon et al.: static-point layout inference plus multipath ray tracing localizes multiple hidden pedestrians at real outdoor T-junctions [IEEE IV]
│ Kim et al.: camera-derived parked-vehicle geometry plus radar refinement handles temporary darting-out occluders in real urban NLOS driving [IEEE/RSJ IROS]
│ Isogawa et al.: C2NLOS transient sinograms — 1D circular confocal scanning with far fewer measurements [ECCV]
│ Rapp et al.: edge-resolved transient imaging — 2.5D room-scale recovery from 45 edge-coded scans [Nature Comm.]
│
2021 ── Nam et al.: real-time diffuse-object NLOS video at 5 fps [Nature Comm.]
│ Plenoptic passive NLOS gains quantitative depth/transverse localization and then life-size 3D LWIR hidden-object localization from scattered thermal light fields [JOSA A / Optics Express]
│ Seidel et al. add angle-and-range 2-D edge-camera reconstruction, while Saunders and Goyal improve strong-background robustness with optimized preconditioning [IEEE TCI / IEEE ICCP]
│ Zheng et al.: ordinary-camera white-light speckle correlation plus a two-step DNN establishes a broadband learned NLOS branch without coherent illumination [Optics Express]
│ Peng et al.: NLOS photography — direct high-resolution hidden-view image synthesis from transients [arXiv]
│
2022 ── CornerRadar: learned RF hint maps enable robust indoor around-corner localization across diverse layouts [PACM IMWUT]
│ Mosaic: diverse curved and planar urban reflectors broaden automotive around-corner radar coverage [MobiSys]
2022 ── Grau et al.: Occlusion Fields — implicit recoverability and self-occlusion-aware hidden meshes [arXiv]
│ Passive THz NLOS uses rough-wall specular components for hidden-human imaging and plenoptic refocusing for 3D localization with room-temperature sensing [IEEE T-THz / Optics Express]
│ Xu et al.: product-convolution expansions turn the shift-variant non-confocal ellipsoidal operator into FFT-accelerated local convolutions [Optics Letters]
│ Wu et al.: an untrained deep decoder is optimized through the passive occluder-aided forward model, avoiding paired training data under strong ambient light [Optics Letters]
│ Nakamura and Takatori: transparent reflective surfaces extend automotive virtual-image localization with parallax and polarization considerations [IEEE ITSC]
│ Li et al.: time-sequential first-photon stamping — detection-aware acquisition reduces photon collection time for active transient NLOS [Optics Letters]
│ Luesia et al.: scattering-media phasor fields — virtual-wave reconstruction remains effective for hidden scenes immersed in thick volumetric scattering [Optics Letters]
│
│ Wei et al.: measured 77 GHz mmWave NLOS 3D imaging establishes mirror-symmetry backprojection and relay-phase compensation [IEEE TGRS]
2023 ── Wang et al.: Signal Superresolution Network — plug-and-play 16× sparse-scan acceleration [CVPR]
│ Boger-Lombard et al.: acoustic daylight interferometry retrieves Green functions from uncontrolled noise for passive around-corner localization [Scientific Reports]
│ Wang et al.: SPIR-Net jointly reconstructs hidden appearance and estimates object position from a single-shot passive speckle pattern, adding learned spatial localization without transient timing [Optics and Lasers in Engineering]
│ Zhu et al.: simulation-trained compressive learning reconstructs 32×32 hidden images from 8×8 transient scans [Physical Review Applied]
│ Li et al.: patterned active single-pixel correlation plus inverse diffraction improves hidden-image recovery [Journal of Optics]
│ Liu et al.: SSCR — mixed-dimensional regularization from 5×5 confocal measurements [CVPR]
│ Li et al.: deep under-scanning reconstruction — learned transient recovery plus physics-aware volumetric inversion from grids as sparse as 8×8 [NeurIPS]
│ Yu et al.: learnable inverse kernel and attention — system-aware transient inversion for hidden intensity and depth [ICCV]
│ Gu et al.: 3D RSD — fast two-stage wave propagation for non-planar relay surfaces [ICCP]
│ Plack et al.: fast differentiable transient rendering — NLOS inverse rendering in minutes [WACV]
│ Choi et al.: self-calibrating differentiable NLOS inverse rendering — jointly optimizes imaging and scene parameters [SIGGRAPH Asia / TOG]
│ Huang et al.: Omni-LOS — joint LOS/NLOS neural transients for near-360° single-position shape [arXiv]
│ Royo et al.: virtual mirrors — higher-order phasor transport and two-corner NLOS [SIGGRAPH / TOG]
│
2024 ── RFlect: practical poles and curved/composite reflectors support hidden-shape mmWave imaging beyond planar-wall assumptions [MobiCom]
│ Xu et al.: rough-relay stochastic scattering and multi-angle double-sparse inversion turn non-ideal wall geometry into recoverable mmWave NLOS structure [IEEE IoT Journal / IEEE TSP]
│ Mehrotra et al.: Hydra explicitly exploits multi-bounce scattering for beyond-field-of-view localization without an environment map [MobiCom]
│ Tosi et al.: 27.4-GHz 5G/mmWave ISAC proves fully NLOS target detection with TDD-artifact-aware CSI processing [IEEE SPAWC]
│ Cui and Trichopoulos: active 270–300 GHz THz imaging treats building surfaces as lossy mirrors and folds multi-reflection paths for centimeter-scale hidden-object geometry and pose [IEEE T-TST]
│ Yasmeen et al.: a measured 5.5 GHz one-bit RIS redirects a monostatic radar around a corridor corner for hidden-human micro-Doppler sensing [IEEE MAPCON]
2024 ── Li et al.: ST-Mamba — state-space temporal modeling and phasor-domain wave supervision for consistent dynamic NLOS video [NeurIPS]
│ He et al. and Yu et al. establish supervised dual-input U-Net and channel-aware GAN passive reconstruction, while Kim and Jang emphasize robustness to changes in occluder and hidden-object geometry [IEEE NNICE / Proc. SPIE]
│ Chandran et al.: learned spotlight selection with an off-the-shelf projector-camera pair turns illumination placement into an end-to-end NLOS sensing variable [WACV]
│ Li and Zhang: USEEN brings passive diffuse-wall NLOS reconstruction to room-scale scenes with 12.2 ms inference [Sensors]
│ Wang et al.: instrument-response deconvolution removes TCSPC timing blur before LCT or f-k reconstruction, converting hardware jitter into an explicit inverse problem [Optics Express]
│ Li et al.: multi-wavelength compressive single-pixel NLOS adds RGB colour recovery at 29% sampling [Applied Physics B]
│ Zhao et al.: liquid-crystal planar angle magnification enlarges the effective relay aperture while correlation-aware sparse scanning cuts acquisition time [Nanophotonics]
│ Wang et al.: vector-optical-field modeling turns illumination angle and polarization into multi-view, low-SNR NLOS measurement degrees of freedom [Laser & Photonics Reviews]
│ Wei et al.: multispectral speckle component separation recovers up to six wavelength channels and validates spectral hidden-object imaging around corners [Optics Express]
│ Pueyo-Ciutad et al.: time-gated polarization — picosecond polarimetric transport reduces the missing cone and recovers directionally ambiguous hidden surfaces [SIGGRAPH Asia]
│ Zhu et al.: computational neuromorphic NLOS tracking — event streams isolate motion-induced speckle changes for direct hidden-target motion estimation [Optics Letters]
│ Sultan et al.: accurate transport modeling — a unified formulation connects transient ToF and occlusion/shadow NLOS [Optics Express]
│ Redo-Sanchez et al.: Dirac-notation framework — backprojection, f-k migration, and phasor fields become comparable operators [Optics Express]
│ Czajkowski and Murray-Bruce: TERI — two doorway edges and one ordinary image enable passive full-colour 3D NLOS [Nature Communications]
│ Czajkowski & Murray-Bruce: Fisher-equalized TERI discretization — depth bins follow edge-resolved measurement sensitivity [ICASSP]
│ Raji & Murray-Bruce: SNLLS and Soft Shadow Diffusion — ordinary soft shadows jointly reveal hidden 3D occluders and 2D radiosity [ICASSP / ECCV]
│ Yue et al.: permutation transient attention — attention and Permute-MLP blocks improve learned transient features [COSI]
│ Zhang et al.: real-time scan-free NLOS — parallel SPAD-array boundary migration reaches 151-fps acquisition and 19-fps reconstruction [APL Photonics]
│ Wang et al.: event-enhanced passive NLOS — asynchronous diffusion-pattern changes and physics-embedded learning reconstruct moving hidden objects [IEEE Sensors Journal]
│ Zhou et al.: white-light ZPF speckle correlation — ambient-light and alignment-robust ordinary-camera reconstruction [Optics & Laser Technology]
│ Huang et al.: comb-calibrated coherent FMCW NLOS — sub-picosecond equivalent timing enables submillimeter 3D imaging and hidden-object vibrometry [Physical Review Letters]
│
│ Cai et al. and Liu et al.: NSIR and RM-CSTV combine echo separation, phase correction, range migration, sparsity, and total variation for measured radar reconstruction [Journal of Radars / National Science Open]
2025 ── Chen et al.: LMS-NLOS couples lightweight multi-scale fusion and attention-guided detail recovery for deployable passive reconstruction [The Visual Computer]
│ Jin et al.: MSPDiff introduces polarization-guided coarse-to-fine diffusion for passive LWIR NLOS reconstruction [The Visual Computer]
│ Lin et al.: CA-SlotNet uses contrast-guided slot routing and physics-aware regularization for classification from temporally truncated photon sequences [IEEE EICARS]
2024 ── Suenobu et al.: physical-optics linearized inverse scattering embeds a known T-junction's multipath Green tensor for simulated hidden-PEC imaging [IEEE ICEAA]
2025 ── Suenobu et al.: the multipath-exploitation formulation is extended to full-wave simulation and measured 2 GHz indoor T-junction imaging [IEEE JSTARS]
2025 ── Fu et al. and Zhou et al.: physics-enhanced and single-shot speckle statistics move steady-state NLOS toward inexpensive white-light and ambient-light operation [Applied Optics / Optics Communications]
│ Zhao et al.: polarization-component extraction separates multiple around-corner targets beyond one optical memory-effect region in a single exposure [Proc. SPIE / OYSS]
│ Chen et al.: Hyper-NLOS introduces wavelength-resolved hyperspectral conditioning, a full-color autoencoder, and spatial–spectral attention for passive hidden-scene reconstruction [Optics Express]
│ Wang et al.: passive hidden-target detection broadens ordinary-camera NLOS from image recovery toward task-oriented hidden-object sensing [Proc. SPIE]
│ Hashemi et al. and Chen et al.: multispectral clutter separation and learned hyperspectral band selection strengthen passive NLOS under realistic backgrounds [IEEE TPAMI / Expert Systems with Applications]
│ Zhang et al.: CMFormer reduces transient-volume memory cost and reaches consumer-GPU real-time reconstruction [Optics and Lasers in Engineering]
2025 ── Shi et al.: fast configurable transient simulation and an open NLOS benchmark [arXiv]
│ Wang et al.: NANO formulates transient NLOS inversion as a noise-conditioned neural operator in continuous function spaces, targeting discretization invariance across sparse/irregular scan grids and photon-starved measurements [arXiv]
│ Shen et al.: MARMOT shifts transient learning toward reusable masked pretraining on TransVerse, with the retained scan subset acting as arbitrary sampling [Visual Intelligence]
│ Li et al.: HAENet extends hyperspectral passive NLOS from full-color autoencoding to spatial–spectral separation and residual attention [Proc. SPIE]
│ Shen et al.: HOLI-1-to-3 combines LOS radiance and NLOS transient fields to complete invisible 3D geometry from one viewpoint [IEEE TPAMI]
│ Klinghoffer et al.: Shoot-Bounce-3D learns to demultiplex single-shot two-bounce single-photon LiDAR for occlusion- and mirror-aware geometry [SIGGRAPH Asia]
│ Scheuble et al.: Transient LASSO fits raw outdoor transient LiDAR with an ambient- and sensor-aware neural scene field [SIGGRAPH Asia]
│ Liu et al.: geometric constraints on hidden surface normals for fast sparse-transient reconstruction [arXiv]
│ Sultan et al.: optimized NUFFT/SFFT sampling — irregular relay scans and flexible hidden-volume grids at FFT-like cost [arXiv]
│ Garcia-Pueyo and Muñoz: inverse phasor fields — diffraction-operator conditioning and a rank-based recoverability metric [Optics Express]
│ Luesia-Lahoz et al.: zero-phase phasor fields — phase zero crossings provide precise and noise-robust hidden depth [ICCP]
│ Czajkowski and Murray-Bruce: visible occluders as opportunistic apertures — passive wide-FoV 3D recovery for extended-depth scenes [EUSIPCO]
│ Zhang et al.: sub-pixel resolving modulation — DMD pixel shifting densifies scan-free relay samples and recovers about 1-cm lateral detail while retaining LCT/virtual-wave/boundary-migration compatibility [Optics Express]
│ Zhang et al.: spatial-correlation scan-free NLOS — a 3D blur-kernel model and correlation resampling recover 2-cm detail at 5 fps from a 16×16 detector [Optics and Lasers in Engineering]
│ Liu et al.: polarization differential correlography — single-shot polarized speckle encoding removes mechanical scanning and achieves millimeter-scale steady-state NLOS reconstruction [Chinese Optics Letters]
│ Liang et al.: HP-CDT — LOS latent representations guide a lightweight hybrid CNN--Transformer for active steady-state hidden-image reconstruction [Signal Processing]
│ Ye et al.: comb-calibrated FMCW tracking — snapshot multi-object position and velocity recovery at 2-mm and 2-mm/s accuracy [Laser & Photonics Reviews]
│ Chen et al.: vector-enhanced sensitive FMCW LiDAR — laser-feedback interferometry and K-domain resampling provide micrometer ranging and millimeter hidden 3D imaging [Journal of Lightwave Technology]
│
│ Lai et al.: HoloRadar reconstructs complete LOS/NLOS 3D scenes with one mobile mmWave radar [NeurIPS]
│ Xu et al.: Bayesian compressive sensing treats imprecise multi-angle relay orientations as latent dictionary parameters [IEEE SPL]
│ Lv et al.: mmWave-only relay-reflector reconstruction removes the LiDAR / known-wall prerequisite for around-corner human sensing [INFOCOM]
│ Luo et al.: multipath-ellipse geometry makes the reflective surface itself an estimated radar state [IEEE TIM]
│ Xue et al. and Jia et al.: unknown L-shaped layouts are jointly inferred with hidden targets from diffraction/reflection paths [IEEE TIM / JSEE]
│ Wu et al. and Xu et al.: binary-accumulation positioning and distributed JPDA extend multipath exploitation toward robust NLOS localization and tracking [IEEE TVT / IGARSS]
│ Alburadi et al.: 228 GHz polarimetric FMCW mapping exploits double-bounce specular paths for measured around-corner imaging [IEEE TGRS]
│ Cai et al.: ACTE-Net and equivariant adaptive-threshold unfolding introduce unsupervised and self-supervised measured mmWave inversion [IEEE TAP / TCI]
│ Miao et al.: adaptive photon-arrival windowing suppresses overwhelming daylight background before TV-regularized transient reconstruction [Optics Express]
│ Mu et al.: wave-propagation and volume-rendering priors enable feed-forward high-speed non-confocal NLOS reconstruction [IEEE TPAMI]
2026 ── Wave-Former: physics-aware wireless shape completion reconstructs complete 3D geometry of fully occluded objects [CVPR]
│ Liu et al.: a dedicated rough-relay mmWave overview consolidates realistic scattering-surface models, multipath-management challenges, and recovery strategies [IEEE AES Magazine]
│ RISE: AoA/AoD multipath enhancement and hierarchical diffusion enable layout reconstruction and object detection from one static radar [CVPR]
│ Zhu et al.: weak-reflector 3-D environmental perception is coupled to path-oriented NLOS moving-target reconstruction [IEEE TAES]
│ Pham et al.: ambiguity-aware particle filtering tracks around-corner radar targets when similar multipath hypotheses yield competing likelihood modes [IEEE TAES]
│ Yasmeen et al.: one-bit RIS quantization produces dual symmetric beams and benchmarks practical around-corner illumination against ideal RIS and metal-reflector baselines [IEEE RadarConf]
│ Yiğitler et al.: Ambient-IoT backscatter fundamental limits — calibration-aware EFIM/CRB analysis identifies which NLOS positioning information survives unreferenced passive anchors and the minimum 2D/3D anchor geometry for joint UE–scatterer identifiability [arXiv]
│ Liaquat et al.: multi-RIS monostatic radar analysis quantifies link-budget and SNR gains for blocked-target detection [IEEE OJVT]
│ Goïcoechea et al.: a single antenna plus programmable RIS reconstructs a full reflection matrix for hidden-scene imaging and moving-target tracking [arXiv]
│ Yu et al.: diffraction-signal utilization estimates building-corner and hidden-target parameters, while Wei et al. correlate multipath ghosts with wall geometry for joint target/layout recovery [IEEE FUSION / EUSIPCO]
│ Yu et al.: an IRS-assisted FDA-MIMO tensor model turns a controllable hidden path into multi-target range/angle localization [Digital Signal Processing]
│ Xie et al. / Yu et al.: covariance-tensor HOSVD and fast coherent/noncoherent angle estimation extend RIS-assisted hidden-target direction finding [IEEE SPL / TAES]
│ Yu et al.: cross-regional NLOS localization moves the multipath-detection branch toward joint-GLRT inference across hidden regions [IEEE IGARSS]
│ Park et al.: reflection-aware camera/radar BEV fusion plus physics-guided ray tracing extends hidden-pedestrian localization to moving ego-vehicle scenarios [ECCV]
2026 ── Yin et al.: Si-SPAD and phase-congruency regularization enable 200 m all-day active NLOS under 94,314 lx sunlight [Optics and Lasers in Engineering]
│ Polarization-encoded spatial multiplexing pushes single-shot speckle around-corner imaging beyond the conventional optical-memory-effect field of view [Optics and Lasers in Engineering]
│ ICT follow-up: industrial 5G/mmWave ISAC adds range--Doppler detection and PHD tracking for reliable fully NLOS intrusion monitoring [ICT 2026]
│ Doğan: laser–acoustic early fusion and LAO-Net extend hidden-human NLOS sensing from localization/reconstruction toward four-class orientation inference [Scientific Reports]
│ Wang et al.: scene-aware audio–visual fusion conditions acoustic spectra on BEV geometry for semantic detection of fully occluded vehicles [IEEE ICASSP]
│ Alakuş and Türkoğlu: wall-mediated chirp echoes and wavelet–acoustic feature fusion enable interpretable nine-class hidden-material recognition [Sensors]
│ Cetin et al.: multichannel wall-mediated echoes extend acoustic NLOS semantics from hidden-material recognition to live-human versus debris discrimination, exposing strong representation–classifier inductive-bias interactions [IEEE Access]
│ Jeon et al.: ASPLE particle filtering and the ARIL/OVAD datasets bring passive acoustic NLOS vehicle localization into road-safety tracking [IEEE T-ITS]
│ Zhai et al.: Biot–Tolstoy–Medwin edge diffraction supplies a physics-aware steering vector for irregular-grid hidden-source beamforming [Measurement]
│ Wang et al.: CorridorLocNet learns real around-corner footstep position from Mel/GCC-PHAT multipath cues with CNN–Conformer cross-attention [JASA]
│ Zhang et al.: monostatic radar--RIS steering-vector decoupling and Root-MUSIC enable simulated NLOS target-angle estimation [Signal Processing]
│ Matsubara et al.: joint hidden thermal-source and environment recovery combines real-image reprojection with supervised synthetic learning [VISAPP]
│ Kozawa et al.: curved vehicle bodies become opportunistic mirrors for camera-only hidden-pedestrian 3D localization [VISAPP]
│ Sheng and Takatori: real-road availability analysis and scale/angle experiments validate stereo vehicle-reflection obstacle localization [IATSS Research]
│ Wei et al.: polarization-encoded spatial multiplexing demultiplexes multiple speckle regions for single-shot around-corner imaging beyond the optical memory-effect range [Optics and Lasers in Engineering]
│ Yang et al.: radar-LiDAR blind-spot complementary fusion supplies measured hidden-target risk cues under complete haul-truck occlusion [Sensors]
│ Spaett et al.: per-pixel SPAD-array timing correction and intrinsic LOS-peak time referencing prevent TCSPC aggregation blur in measured NLOS reconstruction [Optics Express]
│ Yang et al.: Poisson-likelihood Nesterov optimization approaches direct-method speed at 1.56% measured transient sampling [Proc. SPIE / AOMATT]
│ Wang et al.: semantic contours guide relay-wall scan-area and sampling-density reduction for faster active NLOS acquisition [Proc. SPIE / CITA]
2026 ── Zhao et al.: PICL — SPAD-aware denoising cascaded with self-supervised differentiable-physics reconstruction [JOSA A]
│ Ji et al.: a unified scale-modulated, jointly regularized inverse spans through-medium and around-corner NLOS while recovering albedo and depth [Research Square preprint]
│ Ye et al.: NLOSFormer — real-time thermal NLOS through rough surfaces with explicit kernel estimation [ACM TOG]
│ Yin et al.: all-day Si-SPAD NLOS — 200 m reconstruction under 94,314 lx ambient illumination [Optics and Lasers in Engineering]
│ Raji & Murray-Bruce: MDUNet — fast joint 2D/3D passive computational periscopy [WACV]
│ Somasundaram et al.: consumer-LiDAR NLOS via motion-induced aperture sampling [Nature]
│ Lu et al. and Behari et al.: RF geometry reconstruction and low-cost-LiDAR spatial-reasoning benchmarks [CVPR]
│ Lin and Chen: urban-intersection FMCW NLOS perception — chirp-level residual learning suppresses simulated interference before conventional range/angle estimation [Computer Modeling in Engineering & Sciences]
│ Liyanage et al.: multi-antenna micro-Doppler radar preserves cyclostationary rotor cues through cabinets, doors, and walls for experimental NLOS drone detection [IEEE VTC2026-Fall, accepted]
│ Gao et al.: learned LCT — high-frequency-preserving trainable inversion grounded in the original light-cone transform [Physical Review Applied]
│ Sun et al.: TransVID — diffusion-based spatial-temporal interpolation for dynamic transient video [Optics Express]
│ Ling et al.: gradient-coordinated physics-guided training routes conflicting reconstruction, physical-consistency, and sensor-calibration updates instead of collapsing them into one scalar loss [Symmetry]
│ Jin et al.: NLOS-MT combines DeformMamba long-range state-space modeling with windowed-attention U-Net local refinement for active transient reconstruction [ICPR 2026]
│ Wang et al.: diffuse-aware attention encoding for passive NLOS through relay-wall diffusion [Optics Express]
│ Wang et al.: GenPIE — generative geometry priors and differentiable transient transport recover a time-resolved plenoptic representation [SIGGRAPH / ACM TOG]
│ Zhang et al.: D-NeSF — spatiotemporally disentangled neural shadow fields extend two-bounce NLOS stereo reconstruction from static scenes to moving hidden targets [ICGIP / Proceedings of SPIE]
│ Sultan et al.: iterated transient light transport matrices — full relay-wall TLTM recovers hidden-to-hidden time-resolved transport for indirect-shadow analysis, relighting, and dual photography [Nature Communications]
│ Liu et al.: ptychography-enhanced correlography — overlapping indirect speckle probes and geometry-corrected ptychographic phase retrieval recover complex high-frequency steady-state hidden scenes [Advanced Photonics Nexus]
│ Luesia-Lahoz et al.: stereo relay walls — dual virtual apertures reduce the missing cone and recover orientation cues [The Visual Computer]
│ Zeng et al.: compact kilometer-range NLOS — adaptive gating and efficient optics enable daylight 2 fps operation [Optics Express]
│ Yang et al.: MD-NLOS — spectrally filtered model decomposition for sparse real-time reconstruction [iScience]
│
│ Zeng et al.: measured 15 GHz multipath feature fusion recognizes four hidden target classes from path and scattering structure [Journal of Signal Processing]
│ Zhong et al.: physics-guided cross-path contrastive learning recognizes six hidden human activities from UWB radar with only 10% labels [Journal of Radars, Online First]
│ Chen et al.: range-migration and 121 GHz holographic operators embedded in FISTA-style networks extend model-driven NLOS radar to efficient mmWave/sub-THz 3D reconstruction [Journal of Radars / Photonics]
2025 ── Wei et al.: multi-surface waveform deposition — sub-resolution separation of overlapping transient surfaces before LCT reconstruction [Optics Express]
2025 ── Zhou et al.: 10 m passive NLOS — pattern calibration and low-rank background separation move ordinary-camera computational periscopy to long range [Optics Letters]
│ Li et al.: a microscale rough-surface scattering model turns an ordinary rough relay wall into a well-conditioned passive camera, enabling real-time high-resolution and keyhole NLOS recovery [Optica]
2026 ── Liang et al.: HPDI — physics-informed coarse-to-fine reconstruction fused with an implicit data-driven passive pathway [IEEE TCI]
│ Wang et al.: DCEEM makes passive feature selection explicitly light-transport-aware, using Bayesian dynamic channel weighting and hierarchical denoising/refinement to suppress low-SNR channel aliasing before vector quantization [Optics Communications]
│ Yu et al.: enhanced-reflection U-Net reconstruction adds a formally published supervised passive-NLOS encoder–decoder branch [Optics and Precision Engineering]
│ Liu et al.: geometry-constrained NLOS — shape-operator regularization jointly recovers albedo and hidden surface normals [IEEE TVCG]
│ Li et al.: Super-FoV NLOS — translated-PSF spatial encoding reconstructs beyond the physical relay scan region [Photonics Research]
2026 -- Oyama et al.: adaptive spiral relay-wall sampling allocates limited confocal measurements from observed return intensity [IEEE OJSP]
| Liang et al.: cost-effective FMCW interferometry reaches submillimeter ranging and millimeter-scale NLOS imaging [Photonics Research]
| Wang et al.: scanning-free laser reflective tomography demonstrates 3.3-km hidden imaging with centimeter resolution [Opto-Electronic Science]
| Yu et al.: reference-function frequency-domain phase compensation accelerates non-confocal reconstruction [Optics Express]
| Tian et al.: sparse Bayesian transient restoration improves LCT and f-k inputs under photon and hardware limits [Photonics]
| Zhou et al.: polarization-speckle single-pixel modulation opens a scanning-free keyhole NLOS branch [Physical Review Letters]
| Zhang et al.: structure-guided adaptive total variation removes manual passive-NLOS regularization tuning [Optics Express]
| Fu et al.: a diffusion prior restores high-frequency Fourier single-pixel measurements at extreme undersampling [Optics and Lasers in Engineering]
| Chen et al.: multi-attention passive reconstruction handles several simultaneously hidden objects [Frontiers of Computer Science]
| Romanelli et al.: finite-aperture switch-line and Fisher-information analysis quantifies yaw observability [Journal of Imaging]
| Celebi and Turkoglu: real SPAD-TCSPC measurements enable learned hidden-human detection [Sensors]
| Hou et al.: AMPE-NLOS extends active transient semantics to adaptive multi-person 3D body-center and SMPL mesh recovery [Optics Express]
| Xiao et al.: semantic human-pose recovery extends active transients beyond shape reconstruction [Optics and Lasers in Engineering]
| Zhang et al.: dual-variable frequency experts improve robust active inversion [Pattern Recognition]
| Chen et al.: reference-free Canny-guided correction suppresses backprojection artifacts [Optics and Laser Technology]
| Shi et al.: specular-flight-path priors improve non-confocal multi-orientation recovery [Photonics Research]
| Miao et al.: pulse multiplexing computationally exceeds detector timing resolution [Optics and Lasers in Engineering]
| 18 July 2026 reconstruction/pose consistency audit: public-only RF, robotics, renderer, and Quasi-Fresnel records integrated into the survey
| Sun et al.: ordinary RGB relay-wall video enables passive hidden-action recognition and the NLOS-Action benchmark [Neurocomputing]
| Lin et al.: neighborhood-consistency reweighting extends NLOS understanding to robust multi-view clustering [Neurocomputing]
| QSS-Net: recognition-oriented quanta-state-slot modeling advances efficient NLOS semantic classification [FLINS-ISKE / Springer LNCS]
| 18 July 2026 recognition and clustering citation trace: semantic NLOS records integrated into the survey and bibliography
| Li et al.: coherent ultrasound synthetic apertures reach centimeter-scale hidden 3D reconstruction at meter-scale range [Communications Physics]
| Sommer and Katz: knife-edge diffraction enables relay-free passive acoustic NLOS localization [Physical Review Applied]
| 18 July 2026 acoustic citation trace: final-venue ultrasound and edge-diffraction records synchronized across public and survey artifacts
| KAN-Transformer, latent graph regularization, mixture-distribution zero-shot attention, and adaptive artifact cancellation broaden learned and optimization-based active inversion [Optics & Laser Technology / IEEE SPL]
| Long-wave-infrared transport decomposition and commodity CW-ToF scattering mapping extend practical passive and low-cost depth-camera NLOS [Optics and Lasers in Engineering / Photonics Research]
| Visible/LWIR dual-spectral fusion moves passive NLOS toward low-power edge deployment [SPIE NDTA]
| 18 July 2026 citation-gap follow-up: seven final-venue records integrated across public, survey, bibliography, and build artifacts
| Direct CUDA back projection extends active transient reconstruction to arbitrary irregular relay geometry and nonuniform scans [Optics and Lasers in Engineering]
| Reprojection-guided transient recovery advances learned NLOS from fixed sparse grids to fragmented and irregular relay sampling [IEEE JSTSP]
| Flatland provides a real-time 2D transient-and-phasor testbed for controlled end-to-end NLOS analysis [IEEE TCI]
| 18 July 2026 irregular-relay citation trace: three final-venue records integrated across public, survey, bibliography, and build artifacts
| 19 July 2026 phasor/polarization citation trace: inverse-diffraction conditioning and time-gated polarimetric recovery integrated across public, survey, bibliography, and build artifacts
| 18 July 2026 core-citation tracing: forward-citation and modality-expansion audit completed
This repository groups papers by acquisition regime, forward model, inverse algorithm, and sensor modality. The README highlights the newest additions; the website provides a searchable paper explorer and the PDF survey provides the detailed narrative.
If this repository or survey is useful for your research, please cite the original survey:
@article{geng2022recent,
title={Recent Advances on Non-Line-of-Sight Imaging: Conventional Physical Models, Deep Learning, and New Scenes},
author={Geng, Ruixu and Hu, Yang and Chen, Yan},
journal={APSIPA Transactions on Signal and Information Processing},
volume={11},
number={1},
year={2022},
doi={10.1561/116.00000019}
}Corrections, missing papers, and pull requests are welcome. Please include accurate metadata (title, authors, venue/status, year, DOI/arXiv/project links) and a short note explaining why the paper is relevant to NLOS imaging.