2D Object Tracking with a Particle filter and Improved Chamfer Matching

In this paper we propose a parametric object tracking system based on our particle filter with improved chamfer matching. Object tracking is not a trivial work due to the cluttered background, illumination variation and dynamic motion of the target object. To avoid appearance ambiguities, the usually used object descriptors; edge and corner features are better choices than color information. However if the motion of target object is so dynamic and nonlinear, the robustness of object descriptors drastically decrease. Then we adapt a particle filter which is regarded as the best in tracking nonlinear dynamics, with keeping the robustness of object descriptors. Also we propose an effective measurement model, i.e. improved chamfer matching which uses both of edge and corner descriptors to make our result robust. In experimental result, we evaluate our system by comparing Kalman filter and original chamfer matching and show the improved results. C. Oh, M. Z. Islam, J. Kim, S. Kim, C. Lee, 2D Object Tracking with a Particle filter and Improved Chamfer Matching, International Conference on Embedded Systems and Intelligent Technology (ICESIT 2011), pp. 205-208, Feb. 2011