Analysis And Research On Human Abnormal Behavior In Video Surveillance System

In recent years,with the rapid development of Artificial Intelligence and Deep Learning,Machine Learning has become one of the important topics for scientists to study.The abnormal human behavior recognition in video surveillance systems is an important branch of Machine Learning,and it is an indispensable technical means for social security.This algorithm identifies two kinds of human abnormal behaviors in surveillance video.One is the fall behavior of the old who lives alone,the second is the abnormal behavior of pedestrians in the sensitive area.If an abnormality is identified,relevant personnel are notified to make the treatment so that the dangerous situation can be promptly contained.Algorithm study includes the following points.First,an improved GrabCut algorithm is proposed to segment the target and the background of image.The algorithm combines the GrabCut algorithm and the Mean Shift algorithm to solve the problems that the GrabCut algorithm has too many iterations,which results in slow segmentation speed and the changes in the bandwidth of the Mean Shift algorithm affect the segmentation effect.Experiments show that the proposed algorithm can achieve fast segmentation and can maintain image texture and boundary well.When the target and background colors are similar,it also has better segmentation results.Secondly,feature point matching algorithm is proposed to identify the fall behavior of the old.The algorithm computes the Normalized Moment of Inertia of the human body's characteristic points to count the number of matching points of the human body's feature points in the two adjacent images,and compares it with a certain threshold to determine whether the human body has fallen.The experimental results show that the algorithm can effectively distinguish the falling and non-falling behaviors of the human body,and changing the shooting angle or shielding part of the limb has little effect on the recognition result.Finally,an improved dual-line detection algorithm is proposed to identify abnormal human behavior in sensitive areas.The algorithm calculates the number of feature points in the sensitive area and combines the time that the feature point stays in the area to determine whether someone has crossed into the area and makes a stay.The experimental results show that the algorithm has a good recognition effect in the case of single-target or multi-target,and is not affected by the factors such as mutual obstruction and random movement of the target,and it can be used in various scenarios.This algorithm has a good recognition effect on human body's fall behavior and human abnormal behavior in the sensitive area.It can detect abnormal behavior in real time.Compared with other traditional algorithms,the algorithm is less affected by limb movement randomness,and its recognition rate is high and it can be widely used.