Feature matching

What

Point and region feature matching posed as maximum a posteriori inference on a MRF. We go beyond the common 1-to-1 assignment, that is m-to-n multi-assignment. We extend it to multi-frame data association in order to perform tracking of light sources in night-time driving video sequences for vehicle recognition. Also, we apply it to region matching for video co-segmentation.

Video results of papers

Publications

When

2009-12

Who

Jose Carlos Rubio, Joan Serrat