An Heuristic Search based Approach for Moving Objects Tracking

Abstract

Fast and accurate tracking of moving objects in video streams is a critical process in computer vision. This problem can be formulated as exploration problems and thus can be expressed as a search into a state space based representation approach. However, these search problems are hard to solve because they involve search through a high dimensional space. In this paper, we describe an A* heuristic search for computing efficient search through a space of transformations corresponding to the 2D motion of the object, where most promising search alternatives are computed by means of integrating target dynamics into the search process, and ideas from information theory are used to guide the search. The paper includes evaluations with video streams that illustrate the efficiency and suitability for real-time vision tasks on general purpose hardware. Moreover, the computational cost to carry out the tracking task is smaller than real time requirements (40 ms).