Research On Helmet Wearing Detection Method Based On Deep Learning

With the upgrading of computer hardware equipment and the continuous improvement of deep learning technology,the research of deep learning methods for video surveillance has become a hot direction.Using video surveillance technology to identify the wearing of safety helmets of workers in industrial sites can effectively reduce the work pressure of on-site supervisors and ensure the life safety of workers.In this paper,through the improvement of the existing safety helmet detection network,the target detection accuracy and speed are improved.The research content is as follows:(1)Most of the current safety helmet detection models have problems such as low detection accuracy,poor robustness and slow detection speed.The improved XSSD(Xception-Single Shot Multi Box Detector)network model is proposed from the SSD(Single Shot Multi Box Detector)network.The improvement process is to replace the backbone network in the SSD network with the Xception network,which greatly reduces the number of parameters of the network,improves the detection speed.At the same time,the SENet(Squeeze-and-Excitation Networks)module is added to the Xception network,which makes the network pay more attention to the channel characteristics with large amount of information and improves the detection accuracy.Experiments show that compared with the original SSD model,the improved XSSD network model has the detection speed FPS(Frames Per Second)improved from 39 to72,and the average accuracy increased from 71.53% to 75.65%.XSSD network greatly improves the speed of helmet target detection.(2)Although the XSSD network greatly improves the detection speed in safety helmet target detection,the recognition accuracy is not high and the robustness is poor.To solve this problem,this paper proposes a safety helmet wearing detection model LAM-XSSD based on the XSSD network.The improvement process is to integrate the Lightweight Attention Modules(LAM)into the improved XSSD network,so that the improved network model pays more attention to valuable feature.Improved attention to target features leads to higher detection accuracy.Experiments show that the average precision of the LAM-XSSD detection network proposed in this paper is improved to92.90%,and the average missed detection rate is only about 1.20%.The recognition accuracy and robustness of LAM-XSSD network for safety helmet targets are significantly improved,which confirms the superiority of the LAM-XSSD network model.(3)In order to flexibly implement safety helmet wearing detection in the actual scene,a safety helmet wearing detection system is developed based on the improved detection network model.Through the field test,it can quickly and accurately judge the wearing of workers’ safety helmets,and the performance is good.