Research On Identification Method Of Abnormal Driving Behavior Of Automobile Based On Convolutional Neural Network

Efficient identification of drivers’ abnormal driving behaviors is of great significance to prevent traffic accidents.In recent years,with the development of machine learning and artificial intelligence,vehicles are becoming more and more intelligent.In order to effectively regulate the driver’s driving behavior,the camera is installed on the intelligent vehicle to obtain the driver’s relevant information,a convolutional neural network model is built to analyze the data obtained,and the driver’s behavior is effectively classified through the final recognition results.Aiming at the two aspects of "precision and speed",two methods for identifying abnormal driving behaviors are proposed,and the main work contents are as follows:(1)Aiming at the low accuracy of current driving behavior recognition,a recognition model of driver’s abnormal driving behavior based on key sub region feature enhancement is proposed.The model first extracts the driver’s head and hand regions through the sub region extraction model,and then takes the extracted regions and the original images as the input of the model.With the depth residual neural network Res Net-50 as the basic network framework,it effectively identifies ten driving behaviors in the State Farm dataset.The experimental results show that the model has a high recognition accuracy of 91.22% for ten driving behaviors;(2)Aiming at the problem of intra class differences and inter class overlaps of driving actions in low power computing scenarios,a driver abnormal driving behavior recognition model based on multi-scale aggregation module of dual attention mechanism is proposed.Combining the differences between different driving behaviors and the locality of driving actions,the Res Net-50 network is improved as follows:(1)A multi-level feature fusion module is proposed to fuse features at different levels to improve the network model’s perception of feature details.(2)A dual attention mechanism multi-scale aggregation module is proposed,which combines the SE Net channel attention mechanism,non local self attention mechanism and multi-scale aggregation module to form a dual attention mechanism multi-scale aggregation module.Enhance the expression ability of the model in the channel and space,strengthen the local discrimination of driving behavior,and then improve the robustness of the model.The experimental results show that the model can not only classify drivers quickly,but also has high recognition accuracy,with an average of92.98%,and has strong generalization ability.(3)A driver abnormal driving behavior recognition system is designed and implemented.The system is easy to use.The DAMMAM based recognition model of driving abnormal driving behavior is deployed in the system to identify driving behavior efficiently and accurately.This thesis aims at the classification and recognition of drivers’ abnormal driving behaviors,to some extent,to enhance the safety of drivers in the driving process,and provide more reference schemes for the realization of intelligent driving in the future.