An Electronic Fence Application in in Mass Rapid Transit Station Scenarios with the Edge-Computing Approach

This study aims to detect personnel intrusion and personnel classification with a machine vision approach. In the public domain, how to provide a safe environment and convenient services is a topic of great concern to the government, police units and the public. In the past, in the management of intrusion detection on the periphery of the plant area, approaches with constructing infrared sensing and embedding wire to sense the passage of objects have been adopted. Apart from the problem of high construction cost, the hardly distinguishing people, animals or objects often happens. In this study, through image analysis with edge-computing way, the VPU (Video Processing Unit) in NVR (Networks Digital Video Recorder) combined with a machine learning approach is designed to set a virtual no-entry area in the MRT (Mass Rapid Transit) station. When someone enters the no-entry area, the image of person is marked and counted. The results show that the accuracy rate is 90.81% in terms of personnel entering the no-entry area. The main errors occur in personnel duplication marks and multiple duplication marks. In terms of personnel classification, the accuracy rate is 91.32%, and the main errors occur in similar wearing personnel.