Technology And Implementation Of Unsafe Behavior Identification In Tunnel Construction Based On Computer Vision

At present,China is in an important period of continuous improvement of urban infrastructure construction.Many cities are building subways to solve the urban traffic congestion problem to promote the circulation of talents and resources in cities.Industry and resource transfers between cities are also carried out through high-speed rail.During the construction of subways,high-speed railways and other fast-moving lines,the development and excavation of tunnels are inevitably increasing day by day.The safety of tunnel construction is the most important link in the entire tunnel construction process,and it is the basis to ensure the completion of tunnel construction.In recent years,frequent accidents in subways and mine tunnels in China have not only caused huge waste of social resources,but also endangered the lives of construction workers.Against this background,this thesis focuses on the monitoring video of tunnel construction,and uses the function of personnel detection,tracking and behavior identification to get real-time detection results and feedback of unsafe behavior at the tunnel construction site.The main research contents are:(1)To solve the problems of insufficient illumination in the tunnel construction environment and low color contrast of the surveillance video,I took some research and experiments on the histogram equalization algorithm,image enhancement algorithm based on retinal theory,and image enhancement algorithm based on deep learning.It was found that the image enhancement algorithm based on the retina theory is better in visual effects.(2)To solve the problems that the unsafe behavior recognition system of tunnel construction should be detected in real time,I studied the end-to-end YOLO target detection algorithm and determined that the YOLOv3 target detection algorithm is adopted.The Sort series of tracking algorithms are researched and experimented,and it is determined that the DeepSort algorithm is used to complete the real-time tracking of tunnel construction personnel.Finally I build a framework based on the combination of YOLOv3 target detection and DeepSort target tracking algorithm to get the real-time detection and tracking results of tunnel construction workers.(3)Through the research of mainstream action detection algorithms,it is determined that Openpose will be used to detect real-time posture on the tunnel construction personnel,and the ST-GCN will be used to get the position classification results.According to the survey results from the tunnel construction site,the following monitoring functions were identified: the identification of unsafe behavior in fights,unsafe wearing,and cross-border.Based on the research of algorithms,this thesis implements the main algorithms through open source datasets,and generates multiple models of target detection target tracking,and behavior recognition,to complete the monitoring function.At the same time,the monitoring system provides a real-time warning function to the supervisors of the tunnel construction background.In combination with the specific functions of the system,a corresponding database is also designed in this article to record various unsafe behaviors arising from video surveillance.