Driver’s Anger Recognition Method Based On Human-vehicle Environment Information Fusion

With the improvement of people’s living standards,the demand for vehicles is also increasing,resulting in increasingly prominent traffic safety problems.Driver’s driving mood is an important factor affecting traffic safety,among which anger is the most obvious.The anger generated by drivers in the process of driving can easily lead to "road rage",which has a huge impact on the life and property safety of drivers and other traffic participants.Therefore,in order to accurately identify the driver’s anger and reduce road traffic safety hazards,the following research work is carried out in this paper.The simulated driving environment is built and the experimental scheme is designed.In the driving simulation laboratory,a driving simulation environment that meets the expectation of this experiment is built.A multi-source data acquisition system based on UC-win/Road simulation software and Logitech C310-720 P external camera is constructed.The system is used for multi-source heterogeneous data synchronous acquisition.Then the vehicle driving data set and facial expression data set are established.Based on the vehicle driving data set,a combination identification model of Principal Component Analysis(for short,PCA)and improved Artificial Neural Networks(for short,ANN)is constructed.In order to overcome the shortcomings of traditional ANN,the original sigmoid function is abandoned in the input layer and hidden layer,and the Exponential Linear Unit(for short,ELU)function is used as the transfer function of the network.Therefore,a combination identification model of PCA and improved ANN is proposed.The recognition accuracy of the model is 94.67%,the F1 score is 0.8552,and the loss value is 0.0672.The improved Le Net model is constructed based on the driver facial expression data set.To solve the problem that the structure of Le Net network model is relatively simple and can not effectively extract the facial expression features of drivers,this paper introduces the idea of continuous convolution,and improves the structure of Le Net model by forming a convolution pooling unit through two convolutions and one pooling.The recognition accuracy of the model is 95.83%,the F1 score is 0.8837,and the loss value is0.0745.An ensemble learning feature fusion Deep Neural Network(for short,DNN)recognition model is constructed based on vehicle driving data features and driver facial expression features.In this paper,the idea of stacking method in ensemble learning is used in the field of feature fusion,and then a feature level fusion method of multi-source heterogeneous data is proposed to fuse vehicle driving data features and driver facial expression features.The new features are input into the DNN model.The accuracy of the model identification is 98.17%,F1 score is 0.9471 and the loss value is 0.0243.The recognition model established in this study improves the recognition rate of driver anger to a certain extent.The model is a theoretical exploration to reduce the road traffic safety problems caused by anger.It provides a theoretical basis for personalized driving assistant safety system.