Correlogram-Based Perceptual Similarity in Vehicle Probabilistic Tracking

Abstract

This paper addresses the problem of dissimilarity measurement and incorporates an auto-correlogram operator into probabilistic tracking. Perceptual similarity measurement based upon color correlogram is proposed to weigh the candidates and obtain the expectation state vector of tracked vehicles. The object color is represented in perceptually better-organized HSV color space, which produced improvement over the original method that used in the RGB color space. Properly quantized HSV color space improved the efficiency of auto-correlogram computation and robustness to changes in color content and variations in illumination. Experimental results demonstrate that the efficiency of HSV color correlogram in weighing samples is better than that of RGB auto-correlogram and HSV color histogram in traffic scene applications.