Research Of Multi-Moving Object Detection And Tracking For Complex Scenario

Intelligent video surveillance is an emerging research in the field of computervision, which concerns many key technologies such as computer application,pattern recognition, image process, artificial intelligence and mathematics. Now ithas wide applications in medical image analysis, robot navigation, intelligencetransportation, video surveillance and other fields. The research of moving objectdetection and tracking is still one of the most challenging task in the field ofcomputer vision because of the complexity of background, different lightingconditions, complex shape and movement pattern of the object, mutual occlusion ofmany objects and so on.Based on the familiarity and mastery of the relevant digital image processingtheory and technology，this paper makes analysis and comparison of the commonlyused target detection and tracking methods and works on multi-object detectionand tracking for complex scenario.1、About the object detection, the traditional detection and tracking algorithmsare introduced and analyzed. For the problem of light mutation and camera shakein the complicated scene. This paper propose a fast novel method for moving objectdetection drawing lessons from the principle of gray value clustering method.The algorithm divided the pixels into background pixels and foreground pixels,using the clustering method and the pixel-level convergence criteria to establish thebackground and foreground model. The establishment and updating of backgroundmodel adapts to the slow change of lighting and other mutation scene. It can makethe background model approximation to the real background. The establishmentand updating of foreground model improves the transformation from foreground tobackground of mutation scene and helps to develop the background model fast.2、About the target segmentation, use a simple and effective method to binarythe moving target; In the process of morphological processing, use the process ofthe statistical of effective pixels in the sliding box to replace the process of thepixel and mask. In the segmentation process of connected domains, proposed theconcept of equivalent linked list. Use a simple replacement to complete the twoscan algorithm in the second round of scanning. Experimental verification, theimproved algorithm has significantly improved the implement efficiency, meet thesystem real-time requirements. 3、About the target tracking, based on the study of several tracking algorithms,as to the shortcoming of the block, merge and split which occurred in themulti-target tracking, proposed a tracking method based on the KLT feature pointtracking and the Kalman prediction. It can achieve the automatic tracking of themulti-target and have a high tracking accuracy.