Abnormal Behavior Detection For Subway Passengers Based On The Kinect

With the opening of Metro Line 1 and Line 2,Xuzhou has truly entered the Subway Era.With the increasing usage of underground-transportation and the climbing number of passengers,the improvement of the subway station monitoring system has become the top priority.The abnormal behavior of passengers is an important inducement for subway safety accidents,thus monitoring the abnormal behavior of passengers is become definitely necessary.By using the Kinect,the skeleton point data for monitoring and warning the abnormal behavior of subway passengers(falling,illegally jumping over the brake,etc.)was obtained in this work,and a real-time Subway Monitoring System to monitor abnormal behavior of passengers was established.The main research contents and improvement points of this paper are as follows:(1)We detect the falling behavior of passengers by using the method of threshold judgment,as well as template training.Beyond that,we take the use of the Hidden Markov model for training the fall behavior of passengers.Among that,the improved sampling mode for obtaining the data for the rate of decline of spine mid has achieved pretty good results for effectively reducing the impact of data loss on the judgment of falling behavior.In the end,in the simulation tests,the identification rate of fall behavior in all directions is more than 85%,and the misdetection rate is less than 10%.(2)The improved HMM model is used to train and detect the abnormal behavior of passengers’ illegally jumping over the brake.The concept of Decision Tree is introduced and combined with HMM model.By using several joint angles with strong sensitivity to the motion as feature points,the Hidden Markov-Decision Tree model is formed to train and detect the abnormal behavior of passengers.(3)The concept of human skeleton information entropy was introduced to describe the degree of chaos of passengers or passenger flow in order to give early warning to large-scale accidents.The experimental results showing that this method is effective in detecting abnormal behavior of passengers or passenger flow.In the process,we just need to extract the more effective angle of the node that describing the human movement,and obtain the threshold of information entropy in different time periods and different stations.Finally,passengers or passenger flow can be abnormal behavior for early warning.(4)A subway passenger abnormal behavior detection system is constructed by multi-feature fusion.The test results indicate that the system gives a better effect on the detection of the falling and illegally jumping over the brake for the passengers.At the same time,the information entropy of human skeleton can be output for carrying out the accident early warning.The paper has 59 charts,13 tables and 67 references.