| With the determinant of growth quickly and low-carbon and ecologically friendly travel becoming more and more popular,the input of electric motorcycles has been increasing year by year,but the safety risks brought by this have also increased.In order to reduce the injuries suffered by cyclists in accidents,the Road Traffic Safety Law stipulates that helmets must be worn when driving electric motorcycles.However,the detection of whether cyclists wear helmets only through naked eye observation will cause a huge waste of police resources.Helmet recognition of electric vehicle drivers by artificial intelligence algorithm will greatly reduce costs and play a significant role in fostering the growth of smart cities and improving traffic order.This study of the target detection algorithm includes a proposal for an enhanced YOLOv5-based safety helmet wearing detection method for electric motorcycles.The main research contents are as follows:(1)To construct a data set for detecting the wearing condition of safety helmet of electric motorcycle.At present,there is no open data set suitable for this study.Therefore,relevant data were collected by taking screenshots from existing videos,field shooting,web crawler and selecting pictures of electric motorcycles and helmets in VOC data set.After strict screening and cleaning,irrelevant data will be eliminated,and make sense software will be used to mark data according to COCO data set format.The marking categories are divided into two categories: No Helmet or Helmet.The marked data sets are further expanded by Cut Mix data enhancement,which not only increases the diversity of data sets,but also enables the network model to learn more target features.(2)An improved method based on this research is suggested in order to address the issue of existing algorithms’ low accuracy in recognizing complicated traffic roadways and minor target difference.Initially,the K-Means++ clustering method utilized to choose an initial anchor frame that is more appropriate for the goal size in order to increase the prediction frame’s convergence rate.Furthermore,to enhance the extraction of crucial feature values and the model’s ability to detect objects,Coordinate Attention should be incorporated into the backbone network.Lastly,to increase the forecast structure’s convergence speed and regression accuracy,the ?-Io U loss function is introduced.,which provides stronger robustness for small data sets and noise boundary frames.(3)The helmet detection system is designed and implemented based on Python programming language and Py Qt5 framework.The improved YOLOv5 algorithm proposed in this paper is adopted in this system,which can detect not only offline pictures and video data,but also real-time video data captured by cameras.Based on the improved YOLOv5 target detection algorithm,this paper conducted experimental research on the constructed data set.The results show that the improved YOLOv5 model’s m AP achieves 95.1%,which is 7.7% higher than the original YOLOv5’s average accuracy,and the detection speed reaches 63 frames per second.According to the needs of real-time hat use recognition for electric motorcycle drivers in a challenging road environment;The system can greatly reduce police resources and has great practical application and promotion value. |
PDF Full Text Request