Research On Electric Bike Rider Helmet Detection And License Plate Recognition Based On Deep Learnin

With the development of science and technology and the improvement of people’s living standards,the increasing number of electric vehicles has also caused frequent traffic accidents.Wearing safety helmets on electric vehicles can reduce the damage to the head caused by traffic accidents and ensure the safety of the majority of electric vehicle riders.Wearing safety helmets on electric vehicles has become a mandatory requirement for traffic control departments.In order to constrain more and more electric bike riders to wear helmets,a helmet detection and license plate recognition method for electric bike riders based on deep learning is proposed.This method can detect the parts where electric bike riders do not wear helmets and their electric bike license plates,and identify the license plate number of electric bike riders who do not wear helmets,which is used to hold violators accountable.The main work and innovations are summarized as follows:(1)Self-built dataset.By shooting data at traffic intersections and streets,we build a self-built electric bicycle rider’s helmet and license plate detection data set,and then obtain the electric vehicle license plate recognition data set by intercepting the electric vehicle license plate in the above data set and shooting the electric vehicle license plate on campus.Aiming at the problem of insufficient number of self-built data sets and unbalanced samples,the data set is expanded by data enhancement.Finally,the data set is divided and labeled to prepare for the follow-up research.(2)In order to detect the helmet wearing of electric vehicle cyclists more accurately,improved YOLOv5 electric vehicle cyclist helmet and license plate detection method is proposed,which can detect the electric vehicle license plate while detecting that the cyclist does not wear the helmet.The model uses the self-built electric vehicle cyclist helmet and license plate detection data set for training,replaces the GIOU loss function with the CIOU loss function,and replaces the weighted NMS with DIOU＿NMS to enhance the recognition ability of the model for dense riding scenes.At the same time,the ECA attention mechanism is added to the Backone part and the Neck part of the predicted small and medium targets,which improves the recognition rate of the model for small and medium targets.Then,the anchor frame size is re-clustered by K-means algorithm to obtain the anchor frame size suitable for helmet and license plate detection.Finally,Mixup data enhancement is added to enable the model to cope with more complex application scenarios.The experimental results show that the m AP of the improved YOLOv5 electric vehicle cyclist helmet and license plate detection model is 94.95 %,which is 1.46 % higher than the original YOLOv5 model,and4.38 % and 7.35 % higher than the YOLOv4-tiny and Faster RCNN models,respectively.The improved YOLOv5 model can effectively improve the recognition rate of helmet and license plate.YOLOv4-tiny and Faster RCNN models,respectively.The improved YOLOv5 model can effectively improve the recognition rate of helmet and license plate.(3)In order to reduce the number of parameters generated in the license plate recognition process and speed up the detection speed,electric vehicle license plate recognition method based on improved CRNN is proposed.The model uses the self-built electric vehicle license plate recognition data set for training.The 100*32 license plate input size is modified to 94*48 license plate input size,which solves the problem that the two-way electric vehicle license plate number is not easy to identify in the single-line vehicle license plate detection model.Then,part of the conventional convolution of the CNN convolution layer is replaced by a depthwise separable convolution to reduce the number of parameters in the model operation process.Finally,the feature extraction network is redesigned,and the RNN is replaced by 1*7 wide convolution to reduce the running time of the model.Batch normalization and Dropout function are added to prevent model overfitting.The 2,6 and 13 layers of the feature extraction network are pooled and spliced with the 25 layers to enrich the feature information of the output feature map.The output dimension is reduced to 31 to match the number of predicted characters and improve the accuracy of the model for character prediction.Experiments show that the improved CRNN electric vehicle license plate recognition model slightly reduces the recognition rate,the model weight file size is about 1/27 of the original model,and the running speed is 2.7 times of the original model,which is suitable for subsequent deployment on embedded platforms.The electric bicycle rider helmet detection and license plate recognition method based on deep learning is deployed on the Jetson Xavier NX embedded platform and the electric bicycle rider helmet detection and license plate recognition system is built.After testing,the system has a good detection effect on the part of the electric vehicle rider who does not wear the helmet and the electric vehicle license plate,and has a good recognition effect on the detected electric vehicle license plate..