Research And Application Of Multi-target Tracking Algorithm In Intelligent Video Surveillance

With the development of high-performance computer and the popularity of high-quality camera,people's demand for intelligent video analysis is more and more intense.Target tracking has become one of the hotspots in the field of computer vision,and has a very wide application prospect in video surveillance,intelligent transportation,robot navigation and positioning.However,the monitoring videos often contain multiple targets,so the study of multi-target tracking has more research significance and application value,considering the changes in the number of targets and the mutual occlusion between multiple targets.This paper studies the multi-target tracking in the outdoor surveillance videos,and designs a road intelligent monitoring system based on multi-target tracking.The main contributions are as follows:1)Aiming at the problem of target occlusion and intersection,this paper proposes a single target tracking method based on D-STC.The method is based on STC(Spatio-Temporal Context)which introduces the target motion information extracted by direction prediction model based on Markov.That is,the method introduces the constraint of the weighting matrix into the confidence map of the STC tracker context area,using the Hadamard product to obtain a new confidence map,and then get a more accurate target location.The experimental results on the OTB-100 dataset show that the proposed D-STC method has a high accuracy when the targets are occluded and intersected.2)Aiming at the situation that the outdoor surveillance video often contains multiple object,this paper proposes a multi-target tracking method based on D-STC.This method extends the D-STC single target tracking method to multi-target tracking.This method creates a D-STC tracker for each tracking target,and enables them to be independent with each other to achieve multi-target tracking.Firstly,this method uses the moving object detection method based on frame difference and background difference to initialize the tracking target.Secondly,the method uses the multi-objective matching method based on the Hungarian algorithm to associate the tracking target with the detection area.Finally,the method controls the update and disappearance of the D-STC tracker by calculating the similarity between the tracking target and the matching detection area.The experimental results on the PETS2009 dataset show that the method has good real-time and robustness.3)Based on the above research,a prototype system used for road intelligent monitoring has been designed and implemented.The system obtains the trajectory of each moving target by tracking the moving targets,and analyses the movement states of the targets.Then the system can achieves the abnormal behavior(speeding,retrograde)alarm of moving targets and traffic statistics.In the actual road intelligent monitoring,the prototype system has good real-time and robustness.