A Method of Video Abnormal Behavior Detection

Abstract

The invention relates to a video abnormal behavior detection method. The steps of the invention are as follows: firstly, using the three-dimensional scale-invariant feature transformation descriptor to extract the target space-time block feature for the video sequence. Secondly, using spatiotemporal blocks as nodes, the temporal and spatial chaos attributes of the spatiotemporal features of the nodes are calculated, and a spatiotemporal detection model is constructed by combining the target optical flow velocity in the nodes and the Kullback-Leibler distance between each node. The present invention trains different optical flow thresholds for spatio-temporal blocks at different positions, eliminates the influence of the distance between the target and the camera on the optical flow feature extraction, and the combination of local spatio-temporal features and global information has better effects on local and global abnormal behaviors. Good detection effect and improved detection rate.

Description

technical field [0001] The invention belongs to the technical field of image and video processing, and relates to a video abnormal behavior detection method. Background technique [0002] Video detection is one of the most important applications in the field of computer vision, and detecting abnormal events from video sequences has considerable practical significance. Among them, video-based crowd abnormal behavior detection is particularly important. This technology detects crowds in the area and can detect potential dangerous events, thereby improving the response and rescue efficiency of relevant departments. Crowd abnormal event detection is to find abnormal events from surveillance video and send out an alarm. Generally, crowd abnormal events can be divided into local abnormal events and global abnormal events. Local abnormal events refer to the behaviors of some individuals in the crowd that are different from other individuals, such as cycling, roller skating, and ve...

AI Technical Summary

Problems solved by technology

[0003] In recent years, there have been a variety of models for anomaly detection. The classic social dynamic model uses particle flow to approximate crowd flow to overcome the problem that pedestrians are difficult to track. The social dynamic model is established by calculating the interaction force between particles, but this model Lack of social behavior characteristics, and does not reflect the movement behavior characteristics of the crowd

To solve this problem, on the basis of the social dynamic model, social attributes are introduced to express the characteristics of group behavior. Although this kind of method expresses the characteristics of social behavior better, it only uses the local space-time characteristics of the target and ignores the global movement.

Images