Research On Safety Helmet Wearing Detection Algorithm For Construction Site

With the rapid development of my country’s social economy and the further advancement of urbanization,large and small construction sites have spread all over the country.Due to the extensive management of the construction site,the problem of hidden safety hazards has become increasingly prominent,and safety accidents have occurred from time to time.Among all kinds of safety accidents on construction sites,accidents caused by inadequate or damaged safety equipment account for about 80%.In complex construction sites,helmets are a necessary safety device,which can effectively protect the head of the human body and reduce casualties.Therefore,it is an essential construction safety process to supervise the wearing of helmets by the construction site personnel.At present,most of the construction sites use manual supervision or manual video surveillance to detect the wearing of safety helmets by construction workers.However,due to the limited energy of the supervisors and the serious problem of missed inspections,it is no longer sufficient to record whether the safety helmets are worn by the supervisors.Safety production requirements on construction sites.Therefore,this paper designs a safety helmet wearing detection system based on the deep learning target detection algorithm,which can detect the safety helmet wearing of the construction site personnel in real time,and conduct safety supervision on the construction site more efficiently and conveniently.The main work and innovations of this paper are summarized as follows:(1)Investigate multiple construction sites and shoot videos as raw data,and perform data enhancement on the obtained raw data,including noise simulation of different environmental weather,occlusion and cropping,brightness adjustment,multi-angle rotation,etc.Then,use the Label Img tool to label the enhanced data set and create a standard VOC data set,and obtain a total of 10,984 images of the helmet wearing data set,which provides a data basis for the helmet wearing detection model.(2)Propose a YOLOv4-based helmet target detection model YOLOv4-Helmet.First,the CBAM(Convolutional Block Attention Module)attention mechanism module is added to the CSPDarknet53 backbone network to obtain more detailed information of the helmet target and enhance the feature extraction ability of the network;at the same time,for various styles of helmets and small target detection Features,the RFB(Receptive Field Block)module is introduced into the backbone network,and convolution kernels of different sizes and hole convolution are used to obtain receptive fields of different scales;then,referring to the idea of CSPNet to improve the detection speed,CSP is introduced in the feature fusion layer.-N structure,and improve the training strategy to MOSIC-9 at the input end,making it more capable of feature extraction and more robust;finally,the Deepsort tracking algorithm is introduced to solve the problem of false detection and missed detection.The YOLOv4-Helmet helmet detection model proposed in this paper has the characteristics of strong detection ability for small targets,fast model detection speed and high detection accuracy.(3)Based on the YOLOv4-Helmet network model,build a construction site helmet wearing detection system.The system can obtain the video stream of the construction site in real time,analyze the video stream in real time by calling the algorithm model proposed in this paper,and perform real-time alarm on the behavior of not wearing a helmet,and realize the saving of the 15-second alarm video,video screenshots,and video review.,push alarm information and other functions.The field application shows that the system can more efficiently and stably carry out helmet wearing detection and alarm for the construction site personnel.