Research On Target Recognition And Localization Method Based On Multi-camera

With the development of computer vision,image processing and other disciplines,the intelligent development of urban management has been promoted.Intelligent vision detection and tracking technology is also gradually developing.The technology for analyzing and extracting surveillance video information based on computer vision methods is also increasing.Maturity makes visual inspection technology more and more intelligent.With the development of camera technology and the reduction of costs,the limitations of single cameras have gradually become apparent.Single camera monitoring systems have problems such as small observation range,unstable target tracking,and difficult recognition of complex scene occlusions.All need to be taken from the multi-camera monitoring system.The solution has promoted the development of multi-camera surveillance technology.Therefore,the multi-camera-based target recognition and positioning technology has very good research significance and practical application value.Starting from video surveillance,this paper expands the problem of multi-camera target recognition and positioning,and analyzes the three problems of target detection,target tracking,and target positioning.The main work includes the following aspects:(1)Analyze the method of target detection in computer vision,compare several common target recognition algorithms,adopt target recognition and positioning methods based on improved optical flow characteristics,and combine background modeling with optical flow methods to reduce optical flow detection Point to accurately identify the target.(2)Through several common target tracking algorithms,the Kalman filter is used to predict and track the pedestrians that appear,and the Hungarian optimal matching algorithm is used to update the tracking position to achieve continuous matching of moving targets.(3)Aiming at the judgment of the same target being occluded under a certain angle of view in multiple cameras,a target occlusion judgment method combining the spatial position constraint and color characteristics of the target is proposed.First,the homography matrix between multiple cameras is obtained,and then the homography matrix is obtained from multiple cameras.The camera determines the spatial position of the target,and recognizes the same target according to the spatial constraints of the target.Then the color feature of the target is obtained for feature matching.When different target positions overlap and the color feature changes greatly,it is judged that the target occlusion phenomenon occurs.(4)The most important thing for multi-camera target recognition and positioning is the collaborative strategy between multiple cameras.A tracking collaborative strategy is proposed.First,target recognition and tracking are performed on each camera,and then the spatial information and color characteristics of the target are used for Target occlusion recognition,when the target is occluded,the target is occluded and compensated from other perspectives to complete the continuous tracking of the target.(5)Based on the target detection,recognition and tracking algorithm proposed in this paper,build a multi-camera monitoring experiment platform,and design a multi-camera target recognition and positioning system.In this system,artificially given simulation conditions,through accurate spatial models,and the use of collaborative strategies between cameras,to solve the problems of target occlusion recognition and target continuous tracking,and complete the target tracking task under the multi-camera monitoring area.