Research On Algorithms Of Moving Object Detection And Tracking In Complex Video Surveillance

With the development of society and the advancement of science and technology, video surveillance network of cities is developing more and more quickly and has covered various places of human activity. In some sense, Video surveillance has been protecting safety and properties of citizens and greatly reducing crimes. In addition, video surveillance can also be used to manage city traffic and detect traffic jam and accidents for alerting and forecast. In the face of massive data produced by large amounts of video surveillance terminals, the traditional way of manual monitoring can’t meet people’s demand, so the intelligent video surveillance technology is becoming more and more popular.Intelligent video surveillance technique is a branch of computer vision and pattern recognition, which focuses on detection, tracking, recognition and analysis of moving object in video surveillance. Utilizing these techniques, people can quickly obtain the target location, trajectory and behavior of some interesting objects. Moving object detection and tracking is the basis of intelligent video surveillance technology. Although many researches has been done in this field, due to many factors of complex video surveillance environment such as background disturbance and object occlusion, many problems remain to be solved. Against some of the mentioned problems, we proposed advanced algorithms of moving object detection and tracking. The main work is as follows:1. We proposed a moving object detection algorithm based on background modeling and spatio-temporal context, which implicates some potential constraints in video sequences. A new model of spatio-temporal context integrated with background model into a united framework solving a binary optimal question was proposed. Experimental results have proved that this algorithm can effectively deal with the complex video surveillance environment with illumination variation, similar foreground and background, background noise and dynamic background objects such as waving water, fountain and swaying branches. What’s more, the algorithm is very robust with accurate detection and foreground segmentation keeping precise external contour of objects. 2. We proposed an object tracking algorithm based on compressed feature and motion estimation. In order to achieve real-time and accurate tracking results, an efficient compressed feature was introduced. Due to the problem of object tracking drift in online learning algorithms, Kalman filter is utilized for motion estimation which is treated as compensation and correction of native Bayesian classifier. The method adjusting size of tracking video by counting the number of SURF feature points, which considered as information measurement is also discussed. Experimental results validate the algorithm integrating compressed feature in appearance model and motion estimation by Kalman filter is very stable and accurate in complex video surveillance environment.