Abstract

A new method for r eal-time tracking of non-rigid objects seen from a moving cameraisproposed. The central computational module is based on the mean shift iterations and finds the most prob able tar get p osition in the current frame. The dissimilarity between the target model (its c olor distribution) and the target candidates is expr essed by a metric derivedfrom the Bhattacharyya coefficient. The theoretical analysis of the approach shows that it relates to the Bayesian framework while providing a practical, fast and efficient solution. The capability of the tracker to handle in real-time partial occlusions, significant clutter, and target scale variations, is demonstrated for several image sequences. 1

Citations