Research On Driver Distraction Behavior Recognition Based On Deep Learning

With the increase of the tenure of cars,the frequency of traffic accidents is also increasing year by year,which brings severe challenges to traffic safety.Driver's distracted behavior has become one of the main reasons of traffic accidents,thus recognize the driver's distracted behavior accurately can reduce the occurrence of traffic accidents at the source.However,the products that have been widely used in practice have not yet appeared.How to recognize driver's distracted behavior accurately and quickly has become the most essential problem to solve.Traditional recognition methods have few categories and low accuracy.With the rapid development of artificial intelligence-related technologies,various neural network algorithms in the field of deep learning have become more and more mature.Under this background,deep learning method is introduced in this paper to recognize different types of driver's distracted behavior.The specific contents of the study are as follows:Firstly,the five most common distracted behaviors of drivers while driving analysed by the relevant studies are playing with a cell phone,talking on a cell phone,drinking water,reaching for something in the back seat and talking to the first officer.The data set used for this study consists of five distracted behaviors and one normal driving behavior.One part is from the distracted-driver-detection open source data set provided by State Farm,and the other part is a self-made sample data set that simulates the vehicle interior environment.Then the relevant theoretical analysis is discussed of convolutional neural network in the deep learning method,the convolutional neural network model architectures such as VGG16,Inception V3,Res Net50,Efficient Net B0 are established and the simulation comparison experiment is conducted on Tensor Flow for the same data set.Through comparative analysis,the best performance Res Net50 is used as the basic network to improve,the Selu activation function is used to replace the Relu activation function in the original network,and the original Res Net50 network structure is improved.The attention mechanism SE and SE-PRE are added to the residual module.A parallel convolution module is added to the network structure.In order to further improve the model feature extraction ability,the improved Res Net50 network and Efficient Net B0 network are bilinear fusion,that is,the different features extracted from different models were integrated,so the model recognition accuracy can be improved further.Above experimental results can indicate that the proposed scheme based on bilinear fusion convolutional network for driver distracted behavior recognition in this paper has obtained ideal experimental results through multiple experiments,and the detection of driver distracted behavior shows better classification performance.