Color Based Person Tracking with Illumination Compensation by using Multiple Statistics of Scene

Person tracking is an important technology for several applications of video surveillance, robot vision and human interface. Key features of the person tracking are the shape and color. Especially, color information provides important information to distinguish individuals in scenes including multiple persons. However, color information is influenced by illumination color. In this paper, we propose a novel person tracking method based on color tracking with illumination color compensation. The illumination color is estimated from three statistical features obtained from the whole scene. The scene statistics are an average value, a maximum value and "gray-edge" value that means an average of edge component of the scene. Person is tracked as the concentration of registered colors by using the particle filter. In the proposed method, the positions and velocity of points are defined as a system model of the particle filter. Three likelihoods of each particle are calculated by colors compensated by three scene statistics. Then, the maximum likelihood of three likelihoods is selected as the likelihood of the particle. We conducted experiments for evaluation of the proposed method. The results showed the improvement of the accuracy of tracking in comparing with the previous method that compensated color by only average value of the scene.