Research And System Implementation Of Electric Bicycle Helmet Wearing Detection Based On Deep Learning

With the rising number of electric bicycle owners,the problem of accidents associated with them has become more and more serious.According to statistics,among the causes of death in accidents,cranial brain injury is the main cause of death.In order to effectively reduce the accident mortality rate,the Ministry of Public Security in recent years across the country to investigate and correct electric bicycle riders not wearing safety helmets.However,the current supervisory measures taken are human inspection,which cannot achieve long-term effective supervision due to the limited police resources and the large scope of inspection.Therefore,there is an urgent need to use information technology to improve the standard management ability of electric bicycle drivers and riders,to realize the intelligent identification and recording of electric bicycle drivers not wearing helmets,and to establish the association between the behavior and the person who is responsible for the behavior.Based on the above background,this thesis researches the helmet wearing detection of electric bicycle drivers and double-row license plate recognition of electric bicycles based on deep learning theory foundation and applies it to practical engineering.The main work is as follows.1.The safety helmet wearing detection method is investigated,a helmet wearing detection method is designed and an attention-based compact object detection is proposed,a safety helmet wearing detection dataset for electric bicycle drivers is constructed,and the performance of the method is verified on the self-built dataset and the public dataset.2.A light-weight license plate recognition network based on corner point prediction is proposed for an electric bicycle dual-row license plate recognition method,a dual-row license plate dataset for electric bicycles is constructed,and the performance of the method is verified on the self-built dataset and the public dataset.3.Based on the above research,a helmet wear detection system has been designed and implemented.The system consists of a wear detection subsystem and an event management subsystem.The system is capable of identifying and recording the non-wearing of safety helmets by e-bike riders on multiple video streams,identifying the license plate information of e-bikes driven by non-wearing riders and generating non-wearing events.In addition,the system also provides audit and query statistics of non-helmet wearing incidents.