Video-Based Trajectory Analysis And Its Application To Abnormal Events Detection

The video surveillance system uses cameras to acquire videos. It monitors potentially malicious activities through the technology of video analysis. With the rapid progress of technology, the demand of video surveillance systems of modern society increases quickly, reflecting in the continuously need to expand the monitor area and the more and more requirement to improve its function. For the huge amounts of data in video, the user just focuses on some abnormal events. The application value of video monitoring shows in the evolution from forensics after the event to active prevention and monitoring.Intelligent video surveillance system uses computer to analysis data, it detects abnormal events by three steps including the object detection and tracking, the extraction of path and event detection, then it can realize automatic alarm. The objects of video monitoring in this paper mainly focus on two types objects including pedestrians and vehicles, it takes the automatically detection of the abnormal behavior of those two types of objects as the research target, and researches on the technology of intelligent video analysis. The second chapter improves the traditional Mean Shift algorithm on the basis of the existing research of moving object tracking technology. It combines the adaptive feature extraction with the Mean Shift algorithm, aiming at solving the problem of the traditional algorithm’s sensitive to occlusion and illumination. The third chapter studies the path extraction technology, including the extraction and preprocessing of trajectories, clustering and path modeling. This paper uses the method of polynomial fitting to preprocess trajectories, and then uses the improved k-means algorithm for clustering. It improves the initial clustering point selection of the traditional k-means algorithm, making the cluster more effective. The fourth chapter studies the abnormal event detection, the process of detecting is the process of a comparative judgment, and the normal behavior pattern of objects is obtained by training. In the subsequent testing steps, it can be judged whether the object’s behavior is abnormal or not via the comparison of the detected pattern and the normal pattern. What is more, the chapter focuses on the practical applications in the detection of abnormal behavior of pedestrian and vehicle using the technologies of the object tracking, the path extraction and trajectory learning, including converse traffic, abnormal stop, wondering.