Dynamic scene analysis using path and shape coherence

摘要

A new method to find the moving object trajectory from an image sequence using a path and shape coherence constraint is presented. Most motion analysis methods find the correspondence of the tokens (feature points or line segments) or their trajectories from an image sequence based on the smooth variation of the path of each moving token. However, the correspondence is not essentially one-to-one because two sets of tokens extracted from two consecutive frames are not the same. Here, we use the shape smoothness constraint and interpolation correspondence to find the many-to-many mapping of the tokens (line segments). In the experiments, we prove that our method has a faster converging rate and higher corresponding accuracy than previous ones, and it also can find most of the trajectories of occluded and shape variant moving objects.