Efficient Approximation of the Mahalanobis Distance for Tracking with the Kalman Filter

Abstract

In this paper, we address the problem of tracking feature points along image sequences efficiently. Thus, to estimate the undergoing movement we use an approach based on Kalman filtering, which performs the prediction and correction of the features movement in every image frame. Measured data is incorporated by optimizing the global association set built on efficient approximations of the Mahalanobis distance (MD). We analyze the difference between the usage in the tracking results of the original MD formulation and its more efficient approximation, as well as the related computational costs. Experimental results which validate our approach are presented.