Research On The Detection Method Of Helmet Wearing In Complex Posture Based On Deep Learning

Safety helmets are the most basic safety protection device for workers in construction sites with complex environments.However,the quality of domestic workers is uneven,and many workers do not wear safety helmets as required.Therefore,automatic detection of whether workers wear safety helmets at the construction site is of great significance to safe production.In recent years,there have been many researches on the detection of helmet wearing,but most of the research directions aim to improve the recognition rate of helmets under erect conditions,and there are relatively few researches on the detection of helmet wearing under the complex posture of construction workers.This paper combines current research hotspots and specific practical application backgrounds,and introduces a human body posture estimation algorithm for helmet wearing detection research,which can monitor the helmet wearing condition of construction workers.The automatic detection of whether construction workers wear safety helmets under operating conditions reduces the workload of safety officers,and also improves the supervision of enterprises and safety supervision departments,and better realizes the intelligentization of safety production.The main research work is as follows:(1)Establish an image data set oriented to worker operations in construction work scenarios.The data set adopts the PASCAL VOC data set format as the standard,including network collection and on-site images,and is produced using the image labeling software Label Img.(2)Safety helmet wearing detection method based on attitude estimation.In the case of complex postures(such as bending,squatting,leaning back,etc.),a method is proposed to add the posture estimation of the constructor before the helmet detection,and obtain the skeleton point information of the constructor through the improved Open Pose model.Use the three-point positioning method to determine the head area to reduce the detection area,solve the problem of difficult to determine the relative position of the helmet and the human body under the complex posture of the construction personnel,and use the first-order target detection network Retina Net to realize the recognition of the wearing of the helmet,thereby increasing the complexity of the construction personnel The accuracy of the helmet wearing detection in the posture.(3)Safety helmet wearing detection method based on YOLO V4.A detection method for helmet wearing based on head recognition is proposed.Through the cross-validation of facial feature recognition and head recognition,the head area of the construction worker is accurately located,and the problem of the difficult position of the head in the complex posture of the construction worker is solved.Use the YOLO v4 target detection network to detect the wearing of the helmet,and solve the relative positional relationship between the helmet and the human body,so as to judge the wearing of the helmet,so as to solve the problem of low accuracy of the helmet wearing detection under the complex posture of the constructor.