Software Development Of Abnormal Behavior Monitoring System For Prison Based On Video Analysis

The prison is in an extremely important position in the judicial organs of our country.It is of great significance to monitor and deal with the abnormal behaviors in the key areas of the prison in real time to ensure the safety of the prison.Currently,there are many problems in the method that relying on the prison staff to make subjective judgements through monitoring screens,such as low efficiency,high cost and poor accuracy.Therefore,it is of great engineering practical value to develop a high accuracy,real-time and automatic abnormal behavior monitoring system for prison.In this thesis,a software is developed for the management needs of the abnormal behaviors of person loitering,regional intrusion and object abandonment which the prison focuses on.Based on the features of the videos collected by the software in real time,this thesis has developed a pedestrian detection algorithm with high accuracy and real-time performance based on YOLOv3 and designed the loitering behavior detection and regional intrusion detection schemes.Meanwhile,this thesis has designed an abandoned object detection scheme based on the dual background modeling method using Gaussian Mixture Model and performed the GPU parallel optimization by CUDA programming.The software adopts C/S architecture.The server collects and analyzes videos,generates abnormal messages when it detects abnormal behaviors and pushes them to the client in real time to form alarm events.The client presents the video with the alarm events simultaneously,which realizes real-time monitoring and alarming of abnormal behaviors in the prison and meets the needs of the control and disposal requirements of abnormal behaviors in the prison.The test results show that the software developed in this thesis has high accuracy and high real-time performance.In the actual test environment,the system has perfect functions and the accuracy and frame rate have reached 97.52%and 38.07FPS in loitering behavior detection,99.81%and 35.08FPS in regional intrusion detection,95.54%and 249.38FPS in abandoned object detection,which meets the functional and performance requirements of the prison environment.