I know, how meanshift works for tracking objects:
- Histogram of region of interest
- Backprojection image
- calculating center of mass
- iterativ adaption of the ROI's center to the center of mass
But how can the camshift algorithm solve the problem of changing the objects size (e.g. if it moves towards the camera)?
I tried to find some help on Wikipedia, but its written in french and there is no other language available.