Based on work site-oriented security control, this study proposes a security monitoring and early warning based on network video, which is applied to site safety management.The purpose is to detect the helmet wearing behavior on the operation site through the YOLOv4 algorithm, realize the identification frame of the helmet in the monitoring video stream, and alarm the behavior of not wearing the helmet.Labelme software marks the points of interest in the pictures and videos collected on the Internet. After the marking, YOLOv4 algorithm is trained and tested. The overall structure of YOLOv4 is divided into two parts, namely, the main stem feature extraction network and the predicted convolution operation.The main function of the backbone feature extraction network is to extract the features of the target object multiple times, i. e., the process of continuous convolution.The predictive convolution operation is the reuse of a classifier or locator to perform a detection task.They apply the model to multiple locations and scales of the image.Those areas with higher scores can be considered as test results.In addition, relative to other target detection methods, YOLOv4 uses a completely different method, it is a single neural network applied to the whole image, the network divides the image into different areas, thus predict the bounding box and probability of each area, through the prediction probability weighted, this model compared with classifier-based system also has some advantages.After the experiment, the results show that the YOLOv4 identification helmet speed has obvious speed advantages compared with R-CNN and Fast R-CNN recognition algorithms, and realizes the positioning and coordinate display of the safety helmet on the site, that is, the additional target test is successful, realizing the accurate positioning and safety behavior supervision of the construction personnel.After training, the safety supervision system accurately identified the workers wearing yellow safety helmets in the construction site video stream, which achieved the effect of safety supervision.YOLOv4 target recognition algorithm is applied in the safety supervision system in the smart grid development environment, and is combined with the ball machine camera. The supervision process has no influence, does not affect the operation of the staff, improve the on-site operation safety, early warning timeliness, accuracy and comprehensiveness, and then improve the safety level of power construction operation.Under the current normal work intensity of inspectors, it can greatly improve the efficiency of inspectors, reduce the number of mobile cameras and the number of inspection vehicles, providing ideas and direction for the development of the new mode of safety supervision.

 

security monitoring,points of interest,YOLOv4