Research on Recognizing the Wearing Condition of Cycling Helmet Based on YOLOv4
[Purposes]In recent years,governments and social organizations around the world have been increasing their efforts to promote and educate cycling safety.And in the face of the huge number of cy-clists,manpower alone can no longer meet the demand for traffic supervision.Therefore,to improve the safety awareness of cyclists and reduce the occurrence of traffic accidents,a method for identifying the status of the daily cycling helmet wearing is proposed.[Methods]The recognition objects targeted in this study are cycling helmets and two-wheeled vehicles,and the datasets used are all images of cyclists,and the dataset used for recognizing the wearing condition of riding helmets has a total of 1,100 images,which is divided into a ratio of 8:1:1.Then the YOLOv4 algorithm under the Darknet framework is used on the Jetson Xavier NX platform to train and optimize the model of the dataset,which achieves the accu-rate recognition of whether the cyclist wears a helmet or not under the complex scenes with different light intensities.[Findings]The method can achieve a good realization with an average accuracy rate of 80%as well as a low probability of missed detection and false alarms.[Conclusions]The method can realize effective identification of cycling helmets in daily life environment,and can be used in intelligent traffic monitoring,urban management and other fields,which is conducive to improving the level of traffic safety and reducing the incidence of traffic accidents.
NETL
NSTL
万方数据
