A novel probabilistic approach for real time motion segmentation and tracking

Abstract

An adaptive and fully automatic video object tracking scheme is developed on the basis of motion segmentation of the image sequences using a novel probabilistic framework. The inherent idea is to track the moving objects in the current frame and update the frame using a robust Bayesian estimation so that it provides an accurate estimation of the next frame, even when the next frame might be missing. The proposed model uses the homogeneity of image regions based upon probabilistic motion parameters of moving objects in an image to segment them out into video object regions (VOR). Each VOR is modeled as a 4-clique Markov field. Experimental results on the tennis sequence are provided which clearly elucidate that the proposed algorithm is very efficient computationally as well as being accurate and almost real time.