Human Abnormal Behavior Recognition System Based On Deep Learning In Elevator

With the acceleration of urbanization process in China,elevators are being used more and more in high-rise buildings,which brings convenience to people’s life and also has some safety hazards.Because the interior of the elevator is a small and colsed space,it is easy for criminals to use it to commit crimes;elderly people who are physically weak can fall into the elevator if they suddenly fall ill.Therefore,it is very necessary to implement an intelligent detection system for abnormal behaviors of falling,crouching,fighting,aggression in the elevator.When the abnormal behaviors are detected by passengers,the system can promptly alert to prevent accidents.This paper makes an in-depth study on the people counting algorithm and the abnormal behavior recognition algorithm,to establish a high real-time and robust and accurate abnormal behavior recognition system.The main work of this paper is as follows:(1)A people counting model based on convolutional neural network and ridge regression is proposed to calculate the number of people in the elevator.First,the model is improved according to the characteristics of its own data set,and then the improved firely algorithm is used to optimize the ridge parameters,so that the prediction performance of the model is further improved.Finally,the number of people is compared with other algorithms to predict and analyze the results.(2)On the basis of counting the number of people in the elevator,identify the human abnormal behavior.Use a three-dimensional convolutional neural network to identify abnormal behaviro in the case of a single person,change the size of the convolution filter and pooling filter in the time and space dimensions,and then make reasonable selections of parameters such as the learning rate batch value of the network and optimization;then use the double-flow convolutional neural network to identify abnormal behavior in the case of multiple people,and use the advantages of the double-flow convolution network model to design abnormal behavior detection network model,inproved the double-flow model,reduce the amount of parameter and the comlexity of the model,the main idea is to reduce the network level and merge the network before the last fully connected layer of the network.Experiments show that the improved network has a significant improvement in recognition rate.(3)Based on the improved algorithm,the abnormal behavior recognition system in the elevator was designed and built,and the experimental verification was carried out in the real elevator environment,proving that the system could well identify abnormal behaviors in the elevator experimental scene.The paper has 41 charts,8 tables and 76 references.