TremAn Algorithms

Algorithms for TremAn app

Center of motion algorithm

1. Convert frames to gray

2. Compute difference between frames

3. Normalize differences frames (df) from 0 to 1 (here using mat2gray function)

4. Compute center of motion (something like center of mass, but with normalized df) for X and Y coordinates.

   centerOfMotionX = mean(A(:) .* X(:)) / meanA;

   where X are x coordinates of all pixels in A (df).

5. Store this centerOfMotion for every frame.

6. Compute amplitude spectrum using fft for x and y.

7. Get max frequency from average (x and y) spectrum.

Results for testset radim01

• sizeReductionFactor = 0.1 (algorithm works with 1/10 of original resolution)
• works same with sizeReductionFactor = 0.01

video	real freq [Hz] (counting swings)	freq from acc [Hz]	result freq [Hz] (algorithm)
MV1.mp4	12 sw/5 s = 2.4	0*	2.36
MV2.mp4	40 sw/14 s = 2.86	2.86	2.87
MV3.mp4	70 sw/17 s = 4.12	4.06	4.07
MV4.mp4	90 sw/16 s = 5.625	5.48	5.63
MV5.mp4	93 sw/16.5 s = 5.64	5.56	5.57
MV6.mp4	45 sw/17.3 s = 2.54	2.52	2.61

*data are too short and signal is too low, probably does not fit the video.

results for COM algorithm

Current limitations

• Doesn't work for chessboard videos (difference is too sharp )