Delaunay stability

asked 2013-02-11 05:09:29 -0500

Rogeeeer gravatar image

Hi, I'm working on AAM for facedetect, but I have a problem with my Delaunay triangulation : in every videos from youtube it's pretty stable, and the triangles are corresponding between frames. But when I do it, it's not really stable and the triangles are changing (i.e. it's not the same vertices linked each time). It's problematic for the texture warping, if triangles are not the same I can't warp them to find a correct mean texture.

I have a shape with points (vector<point2f>) after adding each point in a subdiv2D (C++, but I tried with the C way) I get the triangle list and print them. My points are always in the same order.

Am I missing something in the principle of the algorithms ? Because I think it's natural that the triangles are not always the same if the shape change...

So is there a way to make this triangulation stable, or do I have to find another solution to warp my texture ?

edit retag flag offensive close merge delete


It may be that they use some averaging/filtering on the tracker position to stabilize it. It's a very common scenario to use a Kalman filter to get rid of detection noise

sammy gravatar imagesammy ( 2013-02-11 06:21:36 -0500 )edit