Particle Filter Object Tracking Based on Color Histogram and Gabor Filter Magnitude

Abstract

Object tracking is a main problem in computer vision, many tracking approaches has been proposed and tested. Color histogram based particle filtering is the most common method used for object tracking [1,2]. Particle filtering is used for its robustness in non-linear and non-Gaussian dynamic state estimation problems and performs well when clutter and occlusions are present, whereas histograms are useful because they have the property that allows changes in the object appearance while they remain the same. However it cannot give a good result if the object and background have the same color, so in order to get a better tracking performance, we introduce a new particle filter tracking method, in which the observation likelihood is calculated using color histogram of the detected object obtained from background subtraction method, combined with Gabor filter features, and we use Box--Muller transformation for state space model. The effectiveness of our approach is verified.