Research on Multiple Abnormal Behaviour Detection Methods for Surveillance Video

Abstract

This paper uses the centroid acceleration of the moving target and the aspect ratio operator of the outer frame of the moving target to identify the human moving target. It uses inter-frame difference method to detect stagnation, uses 3D convolutional neural networks to detect fall, and analyzes the experimental results. Abnormal behaviour detection can assist in case investigation, large-scale activities, and security in key areas, etc., and help ensure social harmony and stability.