Multiple Objects Detection And Tracking In Complex Background

Video surveillance systems have been widely used in recent years. However, most of these systems stay in the stage of video collection. Tasks such as tracking targets in video sequence still rely on system operators. This limitation makes video surveillance systems difficult to find abnormal situations and make early warning in time. The delay of early warning may lead to serious results, for instance, losing life and property of people. Moreover, it also brings negative effects to social public security. Therefore, in recent years, the demand of intelligent video surveillance system increases sharply.In this work, we explore and investigate technologies including stereo vision, moving target detection, and target tracking. The paper proposes improvements to Gaussian mixture model based moving target detection and multiple information based target tracking. By using these improved methods, a real time experimental system is constructed to automatically detect moving targets and tracking these targets in video sequences.This paper consists of four main works:construct the experiment environment of binocular stereo vision system and acquire parameters of this system; improve the update process of Gaussian mixture model by using M-recent frame information and the weight information of Gaussian distribution; propose an effective approach for detecting moving object in medical video sequence; implement of the improved multiple particle filter tracking approach which fuses color information, edge information and depth information.In this paper, we first review the development and current situation of computer vision. Secondly, typical background modeling approaches are introduced. To detect moving objects, the Gaussian mixture background modeling method is adopted. We propose a method to improve the updating process of Gaussian mixture modeling by using M-recent frame information and the weight information of Gaussian distribution. Thirdly, a new approach of moving object detection is developed for medical application. Finally, after briefly introducing some widely used target tracking algorithms, a particle filter tracking approach which combines color feature, edge feature, and depth feature of target is implemented. The proposed approach is applied in this paper to track multiple moving targets.Experimental results demonstrate that the new approach can be used in different computer vision applications. Comparing to other tracking method, the proposed method can better adapt to illumination change and interference of target shadow. As a result, the approach can detect and track multiple targets in video sequences accurately and efficiently.