Vehicle tracking using on-line fusion of color and shape features

Abstract

A real-time on-road vehicle tracking method is presented in this work. The tracker builds statistical models for the target in color and shape feature spaces and continuously evaluates each of the feature spaces by computing the similarity score between the probabilistic distributions of the target and the model. Based on the similarity scores, the final location of the target is determined by fusing the potential locations found in different feature spaces together. The proposed method has been evaluated on real data, illustrating good performance.