Research On Object Occlusion Based On Video Surveillance

The technology of target detection and tracking based on video image sequences has been a hot spot in the field of machine vision research.It has broad application prospect such as military guidance,security construction and intelligent transportation.An intelligent video surveillance system should include: video capture module,image processing module,target detection module,target tracking module and intelligent analysis module,which involves many fields such as pattern recognition,visual analysis and so on.In the video surveillance system,because of the complexity of the background,the target is often part or all of the occlusion.After studying the occlusion problem in target tracking process,this paper proposed method based on the "object permanence" to solve prolonged occlusion problem.The main research contents include:(1)In moving target detection,the background of video surveillance is often not stable,so this paper uses the Gaussian mixture modeling method to extract the prospect of moving object from background.In order to get a relatively complete target image,we use the method of open and close operation to take a connected processing.(2)In terms of moving target tracking,we extract various features of the moving target,including color,edge,texture,histogram and so on.The target tracking process is combined with the trajectory prediction.Focus on solving the problem of tracking the wrong or lost target,caused by occlusion and the missing of single feature.To this end,this paper presents a method based on the "object permanence",and analyzes the relationship between the object and the shelter.It uses various features extracted from the target before the occlusion.The target can be quickly found near the shelter.(3)We carried out a series of experiments to verify the algorithm proposed in this paper.We use "Weizmann","KTH","CAVIAR" standard video library and the recorded video sequences,especially analyze the condition of the light mutation,short time occlusion,long time occlusion,and circular occlusion.Experimental results show that the algorithm can effectively solve the problem of occlusion,and achieve a good tracking effect.