Research On Pedestrian Target Tracking Algorithm Based On Target Split Histogram

With the high intensive and complexity of the social information, the traditional video surveillance cannot meet the demand for real-time processing in emergencies. So, intelligent video surveillance technology has been widely concerned, and now has become one of the world's most concerned fields of the forefront research. As a key technology of intelligent video surveillance, video target tracking mainly solves the problem that the automatic detection and tracking of moving objects in dynamic scenes without human intervention, which has important theoretical significance and wide application prospect. Therefore, this paper mainly research several key issues in multi-object tracking of static video camera scene, are as follows: target matching, data association and nonlinear filter.To reflect the local color feature of the object, a new target matching algorithm is proposed based on the histogram of the sub block of target. In the proposed algorithm, target and template was divided into two blocks(the upper part and lower part), the similarity of the histogram of the two parts are calculated to construct the similarity coefficient function and the similarity between the target and template was obtained by weighted fusion. Experiment result shows that the proposed algorithm has a better performance under the four standards for the similarity of histogram, i.e. correlation coefficient, chi-square, intersection, Bhattacharyya distance.To the problem of multi-object tracking in video with long time occlusion, a multi-object tracking is proposed based on correlation matrix and the histogram matching of the two parts. In the proposed algorithm, the detection responses was obtained by using Gaussian mixture model, the correlation matrix of target and observation was constructed based on the overlapped area of target and observation, On the basis of the correlation matrix, a approach is given for each of the six cases in the association between target and observation. Moreover, to deal with the association problem of multiple targets and single observation, a scale adaptive particle filtering algorithm is proposed based on the histogram of the sub target. Experiment results show that the proposed algorithm can effectively track multiple targets in the case of complex motion, occlusion and target intersection.