Research And Design Of Intelligent Detection System For Construction Workers' Unsafe Behavior

At present,there are many safety accidents in China's construction industry,the situation of production safety is severe,and unsafe behavior of workers is an important cause of safety accidents.Studies have shown that real-time monitoring of workers' unsafe behavior at the construction site can help improve project safety management and reduce safety accidents.It is generally time-consuming,labor-intensive,and inefficient based on the manual supervision of unsafe behaviors by security personnel,but management personnel are also under severe work pressure.The development of an intelligent supervision method of efficient and unsafe behavior that does not rely on manpower meets the current needs of safety management.In view of this,this study proposes an intelligent detection method for common unsafe behaviors of construction workers based on deep learning target detection,and builds a visual operation interface to improve the efficiency of safety management.The main research work includes:First,analyze the principle of deep learning target detection.The basic principles of deep learning are understood,the advantages and disadvantages of traditional target detection algorithms and deep learning target detection algorithms are analyzed,and the two-stage target detection algorithm of deep learning is determined to be the breakthrough point of this research.Combined with the characteristics of construction site monitoring video,Faster R-CNN was selected as the core algorithm for this study.Secondly,the design of the detection system framework.The necessity of real-time detection of four unsafe behaviors was analyzed.Based on the project investigation,the functional requirements of the inspection system were clarified.Combined with the types of unsafe behaviors to be detected and system functional requirements,the overall system architecture is improved,and the overall implementation scheme for intelligent detection of worker unsafe behaviors is proposed.Finally,an intelligent detection system for unsafe behavior is implemented.Through parameter comparison and selection,ResNet-50 is used as a backbone network to extract unsafe behaviors in video.Through the clustering algorithm,the key hyperparameter generation mechanism of the model is optimized,and the model is trained and tested.The model test results show that the improved Faster R-CNN has a MAP value of 0.853 for four types of detection targets,an average detection time of 0.307 s / image,and excellent overall performance.A visual operating system for unsafe behavior detection was created.The system can detect and record the unsafe behavior of workers in real time,issue real-time voice alerts to unsafe behavior subjects,and also provide basic statistical data on unsafe behavior reflecting project safety.The system's running stability was tested,and the problem of multiple overlapping frames generated by the same target was solved.The system test results show that the system can run stably on a computer with high computing power.In theory,this study optimizes the key hyperparameters of Faster R-CNN,which effectively improves the accuracy of target detection.The visual detection system for typical unsafe behaviors constructed in the application can effectively reduce the workload of security personnel and reduce the incidence of safety accidents,and has practical application value for construction safety management.