Detection And Tracking Of Multiple Moving Objects In Intelligent Video Surveillances

The intelligent video surveillance system, which does not require human intervention, automatically analyses the image sequences based on computer vision and video analysis knowledge. It can achieve the objectives of dynamic object location, identification and tracking, then analyze the moving objects' behavior to understand the image content and interpret the image scene. Video surveillance plays an enormous role in the maintenance of social security. Video object detecting and tracking throughout the bottom of video surveillance system is the basement of the following high-level processing like object classification and behavior understanding. Since the 1960s, it has been greatly developed in the military video guidance, security monitoring, traffic control, medical diagnosis and many other applications.This paper presents a set of moving objects detection and tracking algorithm, can carries on the stably tracking in the complex circumstance to the single-object and multi-objects. The paper's main achievements include following several aspects:(1)Two proposed moving objects detection algorithms: the adaptive threshold detection algorithm based on the entropy and the Gaussian Mixture Model base on the Maximum A Posteriori. Aiming at the incomplete detecting question in the traditional methods, the former algorithm gets the adaptive threshold by calculating the entropy energy of the frame difference to segment objects integrally; the latter algorithm based on the Gaussian Mixture Model uses maximum a posteriori theory to separate the foreground and background. Experiments show that these two methods can effectively reduce noise impact and detecting moving object.(2)Adopt the feature based tracking method to track the whole object. In addition to conventional features such as object centroid and area, this paper presents a new feature called Logarithm Illuminance Contrast Statistic (LICS), which represents the object surface feature fully and effectively. And choose an appropriate matching algorithm to track object with these features. Meanwhile, in order to reduce the matching scope of the search, use the Kalman filter to forecast and update the objects' features. It not only reduces the matching error, but also improves the operating efficiency of the system.(3)A sub-block matching algorithm is proposed to handle the occlusion. When occlusion occurred, the whole object will be divided into a number of sub-blocks, and the unoccluded sub-blocks vote on the whole object's state.(4)Combining the above methods, a number of experiments are designed to verify the robustness and correctness of the algorithm.