Moving Objects Detection And Tracking In Complicated Environment

With the rapid development of urbanization, it causes vast amount of problems of issues such as public security, traffic management. To response to the problems with reliable and efficient methods, the study of the Intelligent Video Surveillance or Intelligent Transport System is improving. Technologies of moving objects detection and tracking are the critical support for intelligent system, they make it possible to automatically collect the statistical information of moving targets, so that mining further information like video scene analyzing and behavior understanding.Although a large quantity of technologies in moving targets detection and tracking domain are proposed, there are some difficulties and challenges without good solutions, especially in complex environment. In general, how to improve simultaneously the robust and efficiency of detection and tracking is most severe challenge. Therefore, we proposes a newly adaptive background subtraction for moving objects detection and improving a method of tracking moving targets by integrating MCMC particle filter and Mean Shift.The proposed background subtraction is consisted of three parts:first, we build a sample disparity and a sample consensus statistical mask in the initial segmentation phase. Second, adding a false positives suppression model into the detection method. In addition, it exploits a new update scheme by integrating probabilistic and certain manner. We use the 2012 Change Detection dataset to evaluate the performance of the proposed background subtraction in the experiment. Results demonstrate that the proposed method can suitable for the most complex environment and outperforms the most existing detection methods. Moreover, the proposed tracking targets algorithm absorbs the rapid convergence of Mean Shift and the efficiency of MCMC particle filter for moving targets tracking. And we apply a nearest neighbor data association for multi-targets tracking. In the end, we experiment multi-vehicle tracking with the proposed algorithm in an intersection video from UAV. Results show that our approach performs well.