Human Abnormal Behavior Recognition Based on 3D Residual Convolutional Network

Abstract

In order to ensure the safety of passengers in the elevator, this paper takes the abnormal behavior of passengers in the elevator car as the research object, and proposes an intelligent identification method for abnormal behavior of the elevator car based on 3D residual convolution network, which can improve the accuracy and speed of recognition with the abnormal behavior in the elevator car. This paper combines three-dimensional convolution and residual network, establishes a 3D residual convolution network model, and conducts in-depth research and design of each parameter to identify three abnormal behaviors of elevator cars, such as falling, robbery and fighting. The effects of different video frame input lengths, different parameter optimization functions and whether pre-training on the network recognition rate are compared and analyzed. The experimental results show that when the input video frame length is 16 frames, the Adam optimization function is used for optimization, and the pre-training method can achieve better recognition results.