Maximum entropy method (MEM) for accurate motion tracking

This article summaries a fast, reliable and robust algorithm for real-time tracking of single moving objects in real-world scenes. The proposed tracking algorithm is based on the 2^nd order linear prediction (LP) method and solved by using the maximum entropy method (MEM). The incorporated one-step predictor attempts to predict the centroid of the moving object in the subsequent frame, based on several past centroid measurements of the tracked object. The effectiveness of the proposed recursive predictor-corrector tracking algorithm has been tested with several real-world image sequences containing moving vehicles and people, It has been verified analytically that the proposed algorithm, which employs the MEM is able to yield higher accuracy performance compared to that of the autocorrelation (Yule-Walker) and the covariance method, for a possibly random movement of a single moving object.