Research On The Algorithm For Detecting Abnormal Behavior Of Escalator Based On Video Analysis

As people gradually pay attention to social public safety,the intelligent monitoring system become an important way to ensure social stability and people’s safety.Abnormal behavior detection is the key research content in the field of intelligent monitoring.At present,escalators are located in public places such as supermarkets,shopping malls,and transportation hubs.Safety accidents are prone to occur due to elevator acceleration,elevator failure,objects rolling off,and irregular elevator rides.However,the existing monitoring system often requires a lot of manpower to monitor,and it is difficult to achieve practical and effective detection results.Therefore,the abnormal behaviors that easily occurs on the escalator is detected in this thesis based on the deep learning method.It can improve the early warning mechanisms and improve the monitoring capabilities of the security system.The main content of this thesis is as follows:(1)A set of escalator data sets containing normal and abnormal behaviors is established in this thesis.In view of the sparseness and complexity of abnormal behavior data.An abnormal behavior detection method based on spatiotemporal features is proposed in this thesis.This method reconstructs the normal behavior data through the autoencoder model,uses 3D-CNN and ConvLSTM to learn the spatiotemporal features in video data,and trains a autoencoder model that can characterize normal behavior patterns.Finally,the abnormal behavior detection is performed based on the reconstruction error between the original data and the reconstructed data.This method adopts a lightweight convolutional neural network structure,which can better detect abnormal behavior and realize real-time detection.(2)Aiming at the problem that the generative model is easily interfered by complex background and noise,an abnormal behavior detection method based on foreground targets is proposed.This method reconstructs the normal foreground object to learn the behavior pattern of normal object.In addition,the K-Means algorithm is used to analyze the characteristic distribution of normal behavior patterns,and finally use the one-versus-rest support vector machine to detect abnormal events.Experiments show that the method can effectively improve the accuracy of abnormal behavior detection.(3)Based on the research of abnormal behavior detection,an abnormal behavior detection system for escalators is designed in this thesis.The system can detect abnormal behaviors that occur on the escalator in real time.At the same time,it can detect and track passengers and bulky items,and save important information,which realizes the intelligent monitoring and management of the escalator.