Research On Moving Target Detection And Tracking In Video Surveillance

Moving object detection and tracking is one of the key research topics in the field of computer vision.It has important application value in the fields of intelligent monitoring,medical diagnosis,military guidance,human-computer interaction,unmanned driving and so on.Its application scenario has gradually changed from static scenario to dynamic scenario.Because of the complexity of dynamic scenario,it brings great challenges to moving object detection and tracking.In this paper,the algorithms of moving object detection and tracking in static and dynamic scenes are studied respectively.The main research work of this paper is as follows:(1)Aiming at the problems of ghost and susceptibility to dynamic background in the process of object detection,a multi-scale space based visual background extraction algorithm is proposed.Before the initialization of the background model,the image sequence is transformed by two-level pyramid transform to get three different resolution images,and then target detection and fusion detection results are carried out with different resolution.In the ghost problem,this paper combines the inter-frame information with the second judgment strategy to speed up ghost elimination.Finally,a background complexity metric is proposed,which adaptively adjusts the distance threshold according to the complexity of the background.(2)Aiming at the difficulty of moving target detection in mobile background,a method of moving target detection based on background compensation is proposed.In order to improve the accuracy of background compensation,a background compensation algorithm based on accelerated robust feature matching is selected.In the process of feature matching,two-way matching based on Euclidean distance is used to reduce mismatching.In the parameter estimation,in order to improve the accuracy of parameter estimation,the sequential consistency algorithm and the least square method are combined to solve affine motion parameters,and then completes background motion compensation;Finally,the visual background extraction algorithm is used to detect the target.(3)Aiming at the problem of tracking failure caused by illumination change,occlusion and background interference,an adaptive feature fusion algorithm for target tracking is proposed.By introducing historical model into classifier learning,the algorithm can self-adaptively balance active and passive model learning,so that a more robust model can be obtained in the case of large changes in appearance.In the training stage,the directional gradient histogram and color attribute features of the target are extracted respectively for adaptive feature fusion to improve the feature resolution of the target.Finally,an adaptive model updating strategy is adopted to decide whether to update the tracking model according to the tracking confidence ratio,so as to avoid the contamination of the target model and improve the speed of the algorithm.