Abstract: For the traditional mean shift tracking algorithm,single feature and fixed nuclear window size may result in a track loss when the size and color of targets change.In order to solve this problem,a pedestrian tracking algorithm is proposed based on the multi-feature probability distribution and the mean shift.In the algorithm,first,a target model is constructed based on the color,outline and movement features,and thus the color,edge and movement histogram distributions are obtained.Then,a two-dimensional probability density distribution is created by means of the back-projection of the color and edge histograms,and the color and edge probability distributions are corrected by using the movement information.Moreover,according to the multi-feature weights,the correlation probability distribution of target features is achieved by the adaptive fusion method.Finally,the zero moment of the correlation probability distribution is used to adjust the size of the next tracking window,and by combining the mean shift tracking framework,a normal target tracking is realized.Experimental results indicate that the proposed algorithm is more accurate in extracting target features and can track pedestrians in complex traffic scenes.

Key words: pedestrian tracking, mean shift, histogram distribution, multi-feature fusion, correlation probability distribution