| The frequent occurrence of personnel accidents in the production and operation sites is often caused by the weak safety awareness of the staff and not wearing the safety helmet correctly.In order to enhance the safety awareness of the construction personnel to wear the safety helmet and reduce the incidence of casualties,it is necessary to carry out real-time supervision and detection on the construction personnel.The modern technology of real-time monitoring by camera can effectively reduce the danger caused by not wearing safety helmet correctly.Therefore,in this paper,based on the current domestic and foreign research on the safety helmet detection algorithm,this paper uses deep learning to train a safety helmet detection model,and designs and implements the detection system.This paper makes a deep research in three aspects1)At present,there is little research on safety helmet detection,and there is no open-source database to use.Firstly,we collect data by ourselves,and preprocess the collected data sets by means of mean filtering and median filtering respectively;Secondly,the preprocessed data sets are enhanced to effectively solve the problem of model over fitting caused by less data sets;Finally,Label Img software is used to mark the head,body and safety helmet of construction personnel,which lays the foundation for the establishment of safety helmet detection model.2)After preprocessing,the features of the safety helmet are extracted by using the Darknet-53 neural network,and the position and category of the extracted features are predicted based on the YOLOv3 target detection algorithm.Combined with GIOU curve and loss function,the parameters in the model were optimized and adjusted,and the detection model of safety helmet was established.The model significantly improved the recognition rate of safety helmet.3)Through the combination of the model and camera and other hardware,a system is designed and implemented to detect whether the personnel wear safety helmet in real time after entering the construction site.The system can realize the five modules of login,registration,collection,image detection model and real-time monitoring,which can achieve the purpose of real-time detection and monitoring. |