Research On Abnormal Behavior Analysis Of Pedestrians In Video Surveillance

In intelligent video surveillance,the technology of pedestrian behavior analysis can detect and handle abnormal behaviors in a timely manner,which has great practical value in ensuring social security.This article focuses on how to identify human abnormal behaviors accurately and quickly from video sequences.Based on effective target detection and tracking,human abnormal behaviors can be identified according to the corresponding human characteristics.First of all,this article selects a suitable test video database to provide a reliable guarantee for objective evaluation of algorithm performance.At the same time,the test video is pre-processed to filter out noise during video acquisition or transmission.Secondly,this paper analyzes and summarizes four common moving object extraction methods,compares their advantages and disadvantages,and uses the ViBe+ algorithm with good anti-noise performance and high computing efficiency to detect moving targets.The algorithm extracts the contour of the moving target accurately,and then carries out edge connection,region filling and morphological operations to extract the complete moving target.For problems such as “ghosting” and incomplete target of moving objects in the target detection,the algorithm handles it through propagation suppression,connection area filtering,and other methods.In the tracking of moving targets,the pedestrian's regional characteristics are obtained through target detection,so as to establish a pedestrian area template,and then to track the purpose of the pedestrian through the frame-by-frame pedestrian template matching.In this paper,a pedestrian tracking algorithm based on multi-feature template matching is adopted,and the features of the feature texture,color,and gradient are introduced in detail.Finally,this paper analyzes three different pedestrian behaviors: specific pedestrians crossing the border,pedestrians falling,and pedestrians fighting.This paper uses color recognition and centroid detection methods to alert specific pedestrians across the border.Human body tumbling behavior is recognized by using the characteristics of the aspect ratio of the external circumscribed rectangle of the human body and the characteristics of the center of mass deviation.A human-based fighting behavior identification method is implemented by using a method based on target distance relationship and image optical flow characteristics.