Research on Worksite’s Video Image Safety Helmet Detection Method Based on Cloud-edge Cooperation
Jin Lv, Xiaolong Hao, Min Feng, Zheng Jia · 2022 3rd International Conference on Big Data, Artificial Intelligence and Internet of Things Engineering (ICBAIE) · 2022
In order to ensure the safety of worksites staff, solve the problem that worksite monitoring video transmission occupies many network resources, has high time delay and cannot detect staff not wearing safety helmets in real time, a method for detecting safety helmets in video images of worksites based on cloud-edge cooperation is designed. The YOLOv3 algorithm is used for training the helmet detection model. Using the feature that the cloud-edge cooperation model can reduce processing latency, tasks such as model training and optimization, which are computationally intensive and have low real-time requirements, are handled by the cloud, and tasks such as target detection, which are relatively less computationally intensive and have high real-time requirements, are handled by the edge. Deploy the helmet detection model on the edge node to process the surveillance video images locally, and alarm in time when the behavior of not wearing a helmet is detected; the cloud server receives the surveillance video data and the processing results of the edge, and sends the information to the edge. The information and data are analyzed and calculated, the model is updated and optimized, and then the optimized model is pushed to the edge nodes, so as to realize the data interaction and automatic model optimization of the whole system. The experimental results show that this system can significantly reduce the surveillance video processing delay while maintaining a high recognition rate.