Visual object tracking using particle filter

In this paper, we propose a new observation model with optimizing particle filter framework for visual object tracking in the present of occlusion. Most of the existing algorithms are able to track the object only in predefined and well controlled environment. In computer vision research area, it is challenging task to track when objects get close and circle each others. Some algorithm doesn't even consider the optimization problem. In this work, we develop a robust phase correlation based observation model for particle filter framework and also introduce an optimization technique to improve the performance and accuracy of tracking. Phase correlation provides straight-forward estimation of rigid translational motion between two images. Phase correlation has the advantage that it is not affected by any intensity or contrast differences between two images. Therefore, we apply the phase correlation in particle filter framework for robust tracking. To solve the optimization problem every certain time, we calculate the variance of each weight values to estimate whether there is appeared or not the similar objects around the target object. Based on the estimation results we increase or decrease the number of particles for optimization rather than generate the constant number of particles. We obtained a surprising result with the propose algorithm. Experiments of propose tracker show that our system is robust against cluttered background and optimization problem.