Research On Helmet Detection Technology For Electric Bicycle Riders Based On Deep Learning

E-bike riders wearing safety helmets during the driving process can play a role in preventing injuries and protecting their own lives and the lives of others.The traditional method of judging whether a rider is wearing a helmet by manual observation and giving an early warning is time-consuming and inefficient,so the traffic management department has an urgent need for machine learning-based detection technology for e-bike riders who are not wearing helmets.In this paper,the research on deep learning-based helmet wearing detection method for e-bike riders is carried out,which mainly includes the following contents.A helmet wearing detection method based on an improved Retina Net model is investigated,which consists of two branching networks,one of which is used to detect non-motorized areas such as bicycles and e-bikes that occupy a larger area,and the other branching network is used to detect the head area of pedestrian riders that occupy a smaller area;the feature pyramid structure of the original Retina Net model is improved,and In order to train the improved Retina Net model in this paper and verify the effectiveness of the algorithm,we constructed a dataset containing non-motorized vehicles and pedestrians in multiple scenes by collecting images from actual scenes.74.6% of m AP,which is 7.44 percentage points higher than the original Retina Net model.Considering that the head region of non-motorized pedestrians occupies a small area in the image,and the cyclists wearing helmets and not wearing helmets only show small differences,therefore,in the framework of fine-grained image recognition,this paper investigates a cyclist helmet wearing method based on the improved WS-DAN(Weakly Supervised Data Augmentation Network)model.recognition method,this paper redesigned the network structure of WS-DAN,added a fixed region feature extraction module and incorporated it into the backbone network for enhancing the key discriminative region information of the feature map,and proposed the use of attention mechanism to improve the feature extraction ability of the neural network for the key parts of the feature map.The experimental data based on the test dataset show that the helmet wearing recognition method proposed in this paper can accurately identify whether an e-bike rider is wearing a helmet or not,with a classification accuracy of 94.2% and an F1 Score of 86.23%.