Research On Safety Helmet Wearing Monitoring System Based On Deep Learning

Real-time helmet wearing detection at construction sites has become a difficult challenge for site administrators.It is deeply studied in this paper on the tracking and detection of helmets in the video surveillance system.By analyzing the advantages and disadvantages of present helmet detection algorithms,a deep learning method is proposed to solve the helmet detection problem in surveillance video in accordance with the specific features of application scenarios such as entrance guard places,factory buildings and warehouses,and to change the traditional human supervision mode so as to effectively reduce labor and material cost.In this method,the key point is that the SSD backbone network is used to increase the location information and semantic information of the helmet samples by amplifying the features and copying and pasting them several times,while introducing the deconvolution layer to increase the small pixel layer conv4＿3 feature information for multi-scale feature fusion.Two securities monitoring modes are designed in this method,including the linked access control mode and the dynamic monitoring mode,where he alarm system will be trigged if the helmet is not detected to avoid tragedies.The experiments verify that the proposed method improves the accuracy of helmet detection and can be effectively applied to dynamic helmet detection scenarios in construction sites.