Research On Moving Target Detection In Complex Background

In recent years, moving target detection technology has become an important part of the video surveillance system. Target tracking, identification and analysis are based on the accuracy of the target detection. The topic is one of the key technologies and research priorities of video surveillance systems, especially moving target detection under complex scenes.This dissertation investigates the target detection algorithm based on complex scenes which happened commonly in the process of video. The influence of various complex scenarios on target detection is analyzed. Aimed at the shortcomings of the traditional detection algorithm, the improved moving target detection algorithm is presented and designed.The typical ViBe modeling algorithm has high repetition rate while selecting sample points. To solve this problem, a background modeling algorithm for the choice of sample points area is modified, which select sampling points from 8 neighborhoods to 24 neighborhoods. The algorithm can reduce the repetition rate. ViBe algorithm has some clear disadvantages when it is employed in complex scenes with disturbances and mutations, such as the ghost area. In order to solve this problem, ViBe algorithm and frame difference algorithm are combined to detect moving targets. Whether the background changes suddenly or not is judged by comparing the results which is detected by ViBe algorithm and frame difference. Therefore, the condition for updating background can also be guaranteed. Experiment results show that the detecting algorithm proposed in this dissertation can adapt complex scenes. The problem of the interference caused by background disturbance and scene mutations can be solved effectively.For eliminating the shadow accurately and quickly, the results of several traditional shadow suppression algorithms are investigated and contrasted firstly. Then, a shadow detection algorithm which combines the improved HSV color space and textural feature is proposed in the dissertation. The different color characteristics between shadow and background in the HSV space are analyzed, and the improved HSV detection algorithm is presented. While the textural feature algorithm is employed to detect moving targets, selecting range of the sample points which are applied to initialize from neighborhoods is expanded. At the same time, one pixel disturbance threshold is set to enhance the accuracy of background and anti- disturbance ability. The final shadow detection result is a logical "AND" operation obtained by HSV color space and textural feature. Experiment results show that the shadow can be removed effectively and the real-time performance is also improved.