False Positives in example DrawingInTheAir

[vigneshjathavara]

In the example DrawingInTheAir, even if you don't move the board, there is a detection (at least a very high probability that something is detected even with very slightly similar movement). Hence the rate of false positives is very high.

To reduce the rate of false positives, a check should be done on the distance value from the register IDX_DIST. Hence a threshold can be set on the likelihood of detection, thereby considering any detection with a high IDX_DIST as a no_match.

I'm using a modified classify function like this:

uint16_t Intel_PMT::classify(uint8_t *pattern_vector, int32_t vector_length, uint16_t * dist)
{   
    uint8_t *current_vector = pattern_vector;
    uint8_t index = 0;
    if (vector_length > MaxVectorSize) return -1;    
    for (index = 0; index < (vector_length - 1); index++)
    {
        regWrite16(COMP, current_vector[index]);
    }
    regWrite16( LCOMP , current_vector[vector_length - 1] );
    *dist = regRead16(IDX_DIST);
        
    return  ( regRead16(CAT) & CAT_CATEGORY);    
}

But i guess a CuriePME.getIDX_DIST();
after a call to classify should also work.

[eriknyquist]

Thanks @vigneshjathavara, nice little tip there.
I'll take a look.

[eriknyquist]

Sorry, haven't been able to work on this yet, however you're right, there definitely is a bug there. I should get a chance to come back to this next week