Detecting Abnormal Events In Residential District Based On Intelligent Video Anaysis

The security of the residential area ensures the quality of our life.With the development of computer technology,the intelligent video analysis technology is used in the monitoring system,which has the function of saving time and effort.Moving object detection,tracking and behavior judgment is a very popular research topic in the field of intelligent video analysis.This paper summarizes and compares several classical existing algorithms,and focuses on the three methods for the detection of abnormal events within the residential area,such as wandering detection,climbing detection and the left and move detection.The main work of this paper can be summarized in the following five aspects:(1)Studied the algorithms of moving object detection,such as frame difference method,background difference method and optical flow method,and compare these three algorithms on the public dataset and the private dataset.Because of the importance of pedestrian in the surveillance video,three methods of pedestrian detection are studied.(2)Studied the three classical moving target tracking algorithms,such as the mean shift tracking algorithm,compression tracking algorithm and tracking algorithm based on temporal spatial context,and the three algorithms are compared in the public dataset and private dataset.(3)Based on the above two parts work,presented a method that detect the pedestrian and track in the ROI,and determine whether there is a wandering event based on the difference between the displacement and the distance of the trajectory.Finally,verify the effectiveness of the proposed method.(4)Based on the uniqueness of the climbing posture,proposed a method that detects the pedestrian in the ROI using the joint characteristics of HOG-LBP,and determine whether there is a climb.Finally,verify the effectiveness of the proposed method.(5)Based on the first part prior work,put forward the method that using two Gaussian mixture model with two different rate to extract the objects the motion state was changed in a certain period of time,determine the objects was be left behind or moved according the relationship between object and the background model after excluding the objects was pedestrians.Finally,verify the effectiveness of the proposed method.