Moving Objects Detection And Tracking Algorithms Research In Intelligent Vision Surveillance System

In intelligent visual surveillance system, by applying many advanced algorithm such as image processing, pattern recognition, artificial intelligence and computer vision, and so on, computer can detect, identify and track the changes in the scene automatically and more importantly it can analyse and understand the behavior of moving objects by automatically analysing the image sequences recording by the camera. The system has higher practical value both in the military and civil fields.As two important phases in intelligent visual surveillance system, moving objects detection and tracking technology based on video sequences has been becoming an focus issue in the computer vision domain. The actual monitoring scenes are usually complex and alterable. It is a challenge to detect and track the moving objects accurately, and it is also a basis for the following procedures in intelligent visual surveillance system such as multicamera integration, behavior understanding and description. In this paper, the related algorithms about moving objects detection and tracking are studied based on the relationship of visual information in time and space, in order to improve effect on detection and tracking.Three classic algorithms for moving objects detection are analyzed and studied carefully, and their advantages and drawbacks are compared by experimenting. For the background subtraction method in the static background, this paper focuses on researching foundation and updating methods of mixture Gaussian background model as well as it analyses the background updating strategy, improving robustness of the algorithm adapting scene change. The entire moving objects are obtained by removing shadow and using morphologic method because a number of inanitions and noise lie in the moving region. The results show that the detection algorithm can detect moving objects correctly and quickly and can satisfy the need of the system a-certain extent.For moving objects tracking, the Kalman filter theory is analyzed and the region match algorithm is studied, thus we put forward a region tracking algorithm based on Kalman filter estimate. The Kalman filter estimate model is founded via the characters of objects region, and it estimates moving estate of the objects. Consequently, noise is decreased and search range is reduced, as a result that the efficiency of algorithm is improved. It combines centroid distance with region area difference in order to match objects region, which takes into account the place and size of objects at the same time. Lastly, the prime matching is obtained via the matching rule. The results show that Kalman filter model accords with practical instance and the region match tracking algorithm is able to track moving objects reliably.The occlusion is a difficult problem for tracking algorithm and is a important factor influencing tracking effect at all times. Mean Shift algorithm displays an bad effect for objects tracking in occlusion case by experiment and analyzing it. This paper focuses on achieving objects tracking in occlusion instance by combining Kalman filter with Mean Shift algorithm. Mean Shift algorithm is executed and moving parameters is estimated via Kalman filter in normally tracking objects case, while the subsequent state of objects is estimated using moving parameters and Bhattacharyya coefficient is calculated at the same time in occlusion case. The next frame is taken place occlusion whether or not via using Bhattacharyya Coefficient judging afterwards. It does not stop circulating according to the foregoing method until objects break away from occlusion. The integrated tracking algorithm is able to make up for Mean Shift algorithm a lack of ability via the estimating ability of Kalman filter, which it is incapable of forecasting moving state information of objects in the subsequent frames. The results prove that the integrated tracking algorithm is possessed of a well stability and a strong real-time character in short time of occlusion case.The detection and tracking algorithms in this paper are implemented via programming. The results show that these algorithms possess excellent robustness and accuracy in the little complex background and small change of ray circumstances.