Research And Application On Traffic Statistical Algorithm In Intelligent Video Surveillance

Intelligent video surveillance on traffic statistics is an important application of computer vision. This thesis is based on the monocular video and the vertical shooting mode source as the research object, involved in intelligent video surveillance system the goal of static image segmentation and re cognition, and moving target detection in image sequence, moving target recognition and moving target tracking algorithm research.In the static image segmentation and target recognition, learn a lot about the processing of image segmentation and recognition technology. Combined with the existing head target segmentation image processing technology, the use of the head area more features(such as hair color, contour, etc.) to realize target detection and identification of the head. This thesis use the mean shift algorithm and the combination of the color feature to segment the head target, and use the adaptive Mean shift algorithm segmentation reduce time complexity of the interface f unction relatively Opencv library. After Mean Shift segmentation, combined with hair color characteristics of HSI color space characteristics, color gray features and other techniques to achieve the head target segmentation. Finally, based on the gradient direction information of circle detection Hough transform to achieve recognition head, and whose effectiveness is verified by experiment s.The detection and tracking of head targets in image sequences, different environments lead to different ways of implementation. this thesis select scenes of indoor classroom and outdoor bus for the algorithm research. In the target detection of human body, target detection for scene of indoor classroom in the background to get simple reference model, and choose the background difference method to obtain future target. Outdoor bus due to scene change, illumination and other reasons to select inter- frame difference method, for the inter-frame difference method exists problems, this thesis puts forward the method, the edge information of the image by the combination of the methods of target detection based on color three-frame difference method. Target recognition of the head, on the basis of the characteristics of class round head contour, based on connected domain tightne ss, area, aspect ratio and so on the head prior knowledge to complete target recognition. After the head target recognition, and using the get connected domain of thesis to track achieving goals.Based on statistical new target appear, tracking target temporarily disappear problems, this thesis uses the object-oriented thought to realize automatic variable number of the human body target detection and statistics, and apply it to indoor and outdoor scenes.Finally, through the simulation of the statistic s of the traffic on the above algorithm and experimental analysis, and verify the validity and correctness of the algorithm used in this thesis.