Comparative study of color feature for particle filter based object tracking

Abstract

A comparative study of color feature for object tracking is presented in this paper. The adopted tracking model is the particle filter, which is proven very successful for non-linear and non-Gaussian problems. The color models under study include RGB, HSV and YCbCr. Color quantization is carried out in three components and the histogram is obtained based on the quantized color components for the tracked object. Bhattacharyya distance of object and the predicted position of the object by the particle filter is used to find the posterior probability of particle filter, which is used to update the state of the filter. The evaluation metrics include Displacement Error (DER), Center Distance Measure (CDM). Experimental results show that the object tracking system with the feature selected from HSV color model outperform the system from other two color features.