TRACKING MOVING OBJECTS IN VIDEO BASED ON BENEFIT PART

Abstract

Tracking object in video is important problem of computer vision. Tracking the movement of objects that can be used in security systems, automatic observation, integrated into robots unmanned aerial vehicles of the military, etc. This paper presents an approach that describes the problem in terms of filtering a hidden Markov model using particle filters to filter the random process and combining the computation of confidence scores from gentle adaboost. We suggest tracking the more informative parts (head, body) and omitting the foot part which is the more oscillating part. An InfoPart algorithm is proposes to track the benefit part (head, body) without leg that is more oscillatory. Experimental results show that the average accuracy of InfoPart algorithm is greater than GradNet algorithm.