EmoRecom

Real‑time facial emotion detection with a Flask UI plus location recommendations backed by Google Places.

What it does

• Captures webcam frames, detects faces, and classifies emotions with a trained Xception mini model.
• Streams the camera feed and a scoreboard overlay through Flask endpoints (/video_feed, /video_feed1).
• Stores the dominant emotion so downstream logic can tailor recommendations.
• Finds nearby places (gym, cinema, bar, restaurant, etc.) using Google Places based on the detected emotion.

Requirements

• Python 3.7+
• Access to a webcam and OpenCV video support.
• Pip packages: flask, keras, tensorflow (or tensorflow-cpu), opencv-python, imutils, numpy, googlemaps, googleplaces, requests.
• Model files present:
  • models/face_hyperparams.xml (face detector)
  • models/Xception_mini106.hdf5 (emotion classifier for the Flask app)
  • fer_engine/models/_mini_XCEPTION.106-0.65.hdf5 (emotion classifier for the standalone demo) Placeholders are already in the repo; replace them with the trained weights if needed.

Setup

python3 -m venv .venv
source .venv/bin/activate
pip install flask keras tensorflow opencv-python imutils numpy googlemaps googleplaces requests

Run the Flask app

python server.py

Then open http://127.0.0.1:5000/start in a browser. Routes of note:

• /video_feed – camera stream with bounding box and label
• /video_feed1 – emotion scoreboard stream
• /map – kicks off place recommendations based on the current emotion

Standalone demo (no Flask)

python fer_engine/real_time_classifier.py

This opens a window with the webcam feed and emotion probabilities until you press q.

Notes

• A Google Places API key is required for map suggestions; the code currently reads a hardcoded key in server.py. Replace it with your key before using in production.
• If the camera does not open, ensure another process is not using it and that cv2.VideoCapture(0) matches your device index.