Research On Some Abnormal Event Detection In Intelligent Video Surveillance

Owning to the augment of video contents and wide range of video applications, the application study of video sequences has become a vital research direction of multimedia. The researches on detection, classification and understanding of moving target in video sequences have become hotspots in the worldwide. This thesis focuses on several algorithms of abnormal behavior detection in the video monitoring based on the summary and analysis of the existing fundamental algorithms, such as, loitering detection, human body abnormal detection and detection of abandoned objects.This thesis mainly has five sections:The research background, the significance and the status of the image sequences and the primary content and arrangement of this thesis are introduced in the Introduction section.Some classical algorithms for detecting the moving targets in image sequences are introduced in Chapter2, especially, frame difference, background difference and optical flow method. The feasibility of these classic algorithms is also analyzed, and the results of experiments are provided. The algorithms directly influence the quality of following jobs.In Chapter3some classical algorithms for tracking of movement targets in image sequences are introduced, especially, the Mean-shift, Cam-shift, and Kalman algorithm. They can solve some problems, but still have some limitations. This thesis proposed Mean-shift combined with Kalman algorithm and Mean-shift combined with shape of targets algorithm. These algorithms improve the accuracy.Three algorithms of detecting abnormal behaviors/events in intelligent video surveillances are introduced, i.e. detecting of loitering, abnormal behaviors of human body and abandoned objects. In this thesis, judging in ROI(region of interesting) based on discrete curvature entropy is used for loitering detection, and once the behaviors meet loitering condition, the alarm is raised. For detecting abnormal behaviors of human body, an adaptive abnormal behavior detection method based on isolated point judging is adopted. Experiments demonstrate the efficiency of the proposed method. The traditional algorithm based on codebook is utilized for detecting the abandoned objects. It subtracts between real-time background updating and without updating and alarms when the abandoned objects have been left for some time. These algorithms are based on the detection of moving objects and tracking, on which detecting of abnormal behaviors/events are depended.The summary and the future work are discussed in Chapter5.