Research And Implementation Of Driver Distraction Behavior Detection Based On Lightweight Convolutional Neural Network

While the rise of car ownership and traffic flow brings convenience to people,it also leads to more traffic accidents.A large number of traffic accidents have caused casualties and property losses.The distracted driving behavior of drivers is one of the main causes of traffic accidents.In recent years,the study of distracted driving behavior detection based on convolutional neural network has achieved certain results.However,the parameters of the existing algorithm model are too large and the calculation time is long.The recognition accuracy and speed need to be improved.It is of great practical significance to optimize the accuracy and lightweight of the distracted driving behavior detection algorithm.The specific research contents of this paper are as follows:In view of the small number of distracted driving public data sets and poor generalization,this article collects and builds data sets by itself.Referring to the standards and classification of the distracted driving public data set State Farm,the distracted driving behavior images of 17 drivers of different genders under different lighting conditions and different time periods were collected,and the the collected data set is extended by the data enhancement technology based on traditional and Gaussian noise to improve the robustness and generalization ability of the data set.A distracted driving data set that is closer to the current traffic situation in China has been built.In view of the problem that the speed and accuracy of the existing distracted driving behavior detection cannot be taken into account,a lightweight model based on improved SSD is proposed.First,based on the SSD algorithm,the lightweight network Mobilenetv2 is used to replace the original backbone network for feature extraction,and the unique depth separable convolution and inverse residual modules in the network are used to greatly reduce the number of model parameters,improve the detection speed,and make the model have the conditions to be deployed in the vehicle edge equipment with poor computing power.Secondly,the RFB receptive field expansion module is integrated to improve the ability of model feature extraction,which improves the missed detection problem due to the insufficient information representation ability of the low-level network when the detection target is large.Finally,the CBAM attention module is introduced to make the model pay more attention to the target characteristics,so as to solve the problem of insufficient differentiation between similar actions and further improve the accuracy of model detection.Extensive experimental comparisons shows that the model has both high precision and lightweight.The proposed model can detect and identify six common types of distracted driving behaviors.