Hi!
I need to track two hands that are constantly changing in shape from videos. I have done the basic thresholding. I tried k-means to partition the thresholded pixels into two clusters, and then did contour extraction on each cluster to identify the palm and fingers. But this didn't always work, given that in our video:
- (1) the two hands can sometimes overlap, and
- (2) one hand can be invisible while the other is visible
Given these constraints, could anyone suggest an algorithm that can robustly track two hands simultaneously?
Thanks.