Research On Helmet Detection Algorithm In Complex Scene

According to statistics,most of the safety accidents in Chinese construction industry are caused by falling from high places and hitting objects.As a common and practical personal protection tool,the hard hat can effectively prevent or reduce the head injury caused by falling from high altitude and object impact.But at present,most of the construction sites in China are tested by human eye,which is not only time-consuming and laborious,but also easy to cause mistakes.In view of the current safety management of the construction site,achieving the purpose of real-time monitoring workers’ wearing safety helmets through the implementation of intelligent management of the construction site has important practical significance.By comparing object detection algorithms based on deep learning in recent years,YOLOv5 s algorithm is selected as the basic algorithm in this thesis.Compared with other detection algorithms,YOLOv5 s algorithm has a better trade-off between detection speed and detection accuracy.YOLOv5 s algorithm is the lightest model among YOLOv5 algorithms,with faster detection speed and detection accuracy comparable to other algorithms.However,the construction scene is rather complex,and YOLOv5 s algorithm is prone to omission and false detection in helmet wearing detection.Therefore,further improvement of the algorithm is needed to improve the detection performance of YOLOv5 s algorithm for helmet wearing detection.The main work of this thesis is as follows:In view of the phenomenon that the detection accuracy of helmet wearing detection task is low in complex scenes such as construction site,the loss function of YOLOv5 s algorithm is improved to improve the detection accuracy of helmet wearing detection algorithm.In this thesis,a new?-EIo U loss function is proposed,which replaces CIo U loss function with?-EIo U loss function,so as to improve the positioning accuracy and detection performance of the model.In this thesis,the structure of YOLOv5 s algorithm is improved in view of the omission and false detection a in helmet wearing detection.Firstly,Coord Att attention mechanism is added to the neck of YOLOv5 s algorithm to consider global information,so that the network reassigns more weight information to the helmet,reducing the false detection and missing detection of helmet wearing.Secondly,to solve the problem of inadequate feature fusion in the original backbone network,the residual structure in the backbone network is replaced with Res2 Net Block residual structure,so as to improve the fine-grained fusion ability of YOLOv5 s algorithm,improving the detection ability of small targets.Finally,the K-means++ algorithm is used to cluster the self-made safety helmet data set to obtain the anchor frame information more suitable for safety helmet data set,so as to further improve the performance of safety helmet wearing detection.Experiments and verification are carried out on self-made safety helmet data set in this thesis.Experiments show that the detection accuracy of the improved YOLOv5 s algorithm in this thesis is up to 83.4% m AP and the detection speed is up to 125 FPS.Compared with the original YOLOv5 s algorithm,the m AP is increased by 5.2% and the speed is increased by 6FPS.Experimental results show that the improved algorithm in this thesis not only reduces the false detection and missing detection of helmet wearing,but also improves the detection ability of small targets,so the performance of helmet wearing detection is improved to a certain extent.