Design And Implementation Of Real-time Detection And Alarm System For Safety Helmet Based On Deep Learning

According to the development of society and the progress of science and technology,in the urban mixed traffic,the level of urban motorization has been greatly developed.Various types of public transportation are mixed,and the position that the battery car needs to occupy is relatively weak.According to the preliminary statistics of the Traffic Management Bureau of the Ministry of Public Security of China,the electric bicycle accident caused the death of a driver,about more than 80% of the deaths were caused by injuries to the cranial nerves of the drivers.At present,a number of relevant scientific research results have confirmed that the correct design of wearing a safe driving helmet can reduce the average mortality rate and risk range of various traffic accidents by 60%to 70%.Therefore,in April 2020,the Road Traffic Management Bureau of the Ministry of Public Security once again made important deployment arrangements.Within the scope of domestic urban control this year,the "One Helmet and One Belt" security guard training and education activities will be organized to focus on improving the ordinary people’s daily life.Wear a safety helmet and carry a green safety belt for traffic prevention and safety awareness.Therefore,it has a strong practical significance to detect the wearing of safety helmets for electric bicycle riders in a mixed traffic environment.At present,due to the new release of laws and regulations,there are few researches related to safety helmet wearing detection,related data sets still need to be improved,and related detection systems are still lacking in pertinence.In response to the needs of the traffic police brigade in Beibei District,Chongqing,this article will improve the detection and alarm system based on the detection and identification of electric bicycle riders and the detection and identification of safety helmets.The main tasks completed in the thesis include:(1)A Center Net-based detection algorithm for electric bicycle cyclists is implemented.First of all,due to the lack of data set,the collection of the battery car cyclist detection data set was completed and the data was enhanced.Secondly,using the FPN(feature pyramid networks,feature pyramid)method for reference,the highdimensional feature map obtained in the previous sampling and the low-dimensional feature map obtained in the next sampling are respectively fused,and only the last feature map is used for prediction.The network structure has been improved.Finally,the output and post-processing are improved by setting the heatmap channel for predicting the target center point offset and the branch output of the target size and expanding the neighborhood of the maximum value.(2)Implemented a helmet-wearing recognition algorithm based on HOI.The traditional prediction problem of interaction points and interaction vectors is transformed into the detection problem of the interaction frame composed of the center point of the person and the center point of the object to determine whether there is a "wearing" relationship between the head and shoulders and the helmet,and to further determine the riding of the battery car Whether the personnel wear a helmet.This detection will be implemented under the same Center Net network as the cyclist detection.(3)This paper combines the data detection model we have mentioned above and built,and designed and implemented a web-based B/S(Browserad /server,combined structure of web browser and server)and C/S(Client/ Server,user terminal and server combined structure),a battery car cyclist safety helmet recognition and alarm system.The front-end module of the system is written based on Py Qt.It captures the images of the battery bike riders through the camera at the red light intersection,performs target detection,target recognition and alarms for not wearing a safety helmet,and can record illegal behaviors to the back-end server.The back-end management module is written by the Flask framework and applies the Admin LTE framework.The administrator can retrieve illegal behavior images and view user login logs in the back-end.An application network of on-site front desk module and rear management module is formed,which strengthens the safety and reliability of the entire system.This paper improves the existing target detection algorithm,and solves the problem of improving the accuracy of the safety helmet detection task.While improving the algorithm,it also satisfies the efficiency of the program operation;on the basis of the algorithm in this paper,the safety of the electric bicycle is carried out.The construction of the helmet real-time detection and alarm system has certain practical significance.