Detection And Motion Patterns Analysis Of Object In Multiple Camera System

In recent years, with the development of artificial intelligence, computer vision researchers draw much attention to the research on target motion analysis. Traditional video surveillance systems only contain moving object detection and tracking, and they can’t meet the needs of the recent application. This paper is a design that target motion analysis system based on multi-camera networks. The main contents include: target segmentation, target tracking in single camera and multi-camera, as well as the clustering and analysis of trajectories.First of all, in allusion to different light intensity, changing illumination situation and the shadow problem of side lighting scene, this paper studies a biomimetic visual target segmentation method based on interactive information. The principle of this method is that based on the foreground pixel segmentation using recent studied conventional biological vision models, and then establishes a bionic target segmentation model of computer vision. Model includes three layers: primary processing layer achieves the image to remove noise and other primary processing; access layer completes image processing in various functional modules, independently; visual layer by processing the results ultimately realizes the interactive summary of each channel’s output.Secondly, in allusion to the main sights tracking image sequences containing switching case, it establishes a multi-camera environment target tracking system. In the first place, it needs to select the main camera which sight is the best, gets the main sights image sequences, and by WLMC tracking algorithm based on HSV color features, calculates coordinates of the targets’ to obtain the moving trajectories in the image planes, then using the camera calibration techniques obtains the correspondence between the image plane and the ground plane, ultimately making use of multi-camera data fusion, determines the actual target trajectories.Finally, it studies the cluster and modeling of moving target trajectories, and discusses the movement of a person as the target object. In daily life for people walking normally not random but regular motion, firstly using the K-Means algorithm to cluster trajectories, obtains motion patterns, and then using the GMM model to learn the patterns, predicts the current movement.