I use hierarchical meshes to compute motion vectors on a video in order to
realize motion tracking
As motion on a video is discontinue, my meshes can't track the video for a
long time.
So i would like to know if there is anyone who knows techniques that
- avoid meshes to loose the "solution",
- can determine the point of discontinuities and separate the mesh into
submeshes to continue the tracking.
- can robustufy my approach
Moreover , i use a multi grid approach to estimate motion vectors betwwen 2
images t and t+1, is there better algorithm than
this one? Is there good choice of initialization of the process to speed
convergence?
thanks.
nathalie Laurent