The Research On Moving Target Tracking System On Vision

Moving object detection and tracking is one of main projects of computer vision research,which fuses many advanced technology like image processing, pattern recognition, automaticcontrol, artificial intelligence and computer science and other fields and it has widespreadapplication in military guidance, smart surveillance, medical diagnosis, traffic monitoring androbot field, thus the research of this project has great significance in the academic andapplication value. In this dissertation, the key technology of this project like image preprocessing,object detection and object tracking will be presented and the following is the main work:Firstly, the present study status and methods in domestic and overseas of moving objectdetection and tracking are summarized and the advantage and disadvantage of each method arediscussed, which establishes the important research foundation for this project.Secondly, the basic principles of average filtering, median filtering,mathematicsmorphology and connectivity testing are introduced and the advantage and disadvantage of eachmethod are analyzed through contrast experiments.Thirdly, two methods of image segmentation like the histogram segmentation algorithm andthe OTSU threshold segmentation algorithm are presented. Then three moving object detectionmethods like the inter-frame difference algorithm, the background differencing algorithm and theoptical flow method are elaborated. The method of object detection using background estimationwhich is based on GMM background updates algorithm is achieved and the experimental resultsshow that the detected object is complete.Finally, the principles of Mean Shift algorithm and Kalman filter are elaborated. Then thisthesis proposes a new approach combined Mean Shift tracking algorithm and Kalman filterwhich is applied to predict the motion information of object is introduced and the tracking theobject which has changing size is achieved. Then a new approach which is applied to trackmultiple objects combined blob modeling and modified Mean Shift tracking algorithm ispresented and the experimental results show that the approach is effective and robust.