Contact: [email protected]
Faster computation of HOG features for object tracking using Matlab and OpenCV with a mex interface. For a deteails about the study, please see this file. Any feedback is immensely appreciated!
The original tracking framework used Matlab's implementation of HOG features, which resulted in FPS in the range of 0.35 to 1, whereas optimized implementation achieves FPS that range between 1.39 and 12.89. This was accomplished with the help of OpenCV's library and Matlab's mex interface.
Matlab and OpenCV: https://www.mathworks.com/discovery/matlab-opencv.html
First, call the class constructor:
hogObj = HOGDescriptor(size(image),'CellSize',CellSize,'BlockSize',BlockSize,'BlockOverlap',BlockOverlap);
Then, compute the features with:
featFst = compute(hogObj,image);
When done, release the object:
release(hogObj);