Research On Video-based Detection And Tracking Method Of Intelligent Surveillance

This thesis was focused on the technique of detection and tracking of the Intelligent Surveillance System. Detection and tracking of moving objects in Intelligent Surveillance System was one of the most important research topics in the domain of computer vision and digital image processing, it was also an active field which has interested many researchers in recent years. This paper researched into the moving detection and tracking, the further study was done in this paper and the new methods were brought forward. Moreover, experiments were implemented to demonstrate the validity. The main contents of the study included such aspects as follows:1. Based on frame difference, the motion detection algorithm with the combination of the optical flow method and time filtering was proposed. On assumption of the condition that the objects moved concurrently in the period of time, this method could real-time detect the motion region within the complex background. The algorithm proposed in the thesis was well robust; meanwhile, it could obtain the real-time and robust motion regions with the combination of gradient difference image and time filter, without any priori knowledge such as the size and shape of the objects in the background.2. Extract shadow areas based the conclusion that the brightness of shadow areas was lower than that of the corresponding background. Firstly, divide the detected results into the two kinds of parts that high-brightness and low-brightness, the parts of high-brightness were constructed of the moving objects completely, while the parts of low-brightness were constructed of the low brightness parts of the moving objects and the shadow areas. Secondly, classify the parts of low-brightness by the way of partitioning the shadow areas and low brightness parts of the objects with multi-level thresholds. Finally, combine the parts of high-brightness and the low-brightness, the accurate areas of shadow could be obtained through morphological de-noise processing. The practice proved that the method could eliminate the shadow of moving objects, real-timely and efficiently.3. Tracking for the moving objects based on particle filtering. When real time tracking for the moving objects, the method based on Kalman filtering usually was an efficient method, however, there would be a situation that the objects near to each others were recognized as one object. Errors would occur when objects matching. Hence, the method of replacing Kalman estimation by partical filter was presented. The practice proved that the method could well track the moving objects meanwhile basically achieved real-time requirement of 60msps.