Study On The Recognition Method Of Driving Anger Considering Drivers’ Language

With the continuous advancement of urbanization and modernization,the total number of motor vehicles in China has been increasing year by year.In the urban road traffic environment,due to traffic congestion and the bad behavior of some traffic participants,drivers often generate anger,which leads to traffic accidents.In order to reduce traffic safety hazards and accurately identify drivers’ anger emotions,this paper studies the driving anger emotion recognition method considering drivers’ language factors.The specific research content is as follows:Through a questionnaire survey,driving habits and language habits of drivers are collected to obtain driving behaviors that cause angry driving.At the same time,the commonly used anger keywords are counted.The anger keywords are quantitatively scored from three aspects: anger degree,occurrence frequency and influence degree on driving safety.By fuzzy comprehensive evaluation,anger keywords are empowered,anger words are graded and anger language evaluation system is established.A simulated driving environment was built,a multi-source data acquisition system(including UC-win/Road simulation software and external camera)was constructed by using the simulation driving laboratory,and multi-source synchronous data acquisition(language information,vehicle driving information and face image)was carried out with the help of artificial assistance.Based on driving data and drivers’ facial features,two anger emotion recognition models are established and their performance is verified and analyzed.For the anger recognition model based on vehicle driving data,Principal Component Analysis(PCA)was used to extract the principal components from the data,and then vehicle driving data set was established.An improved Radial Basis Function Neural Network(RBF)classification and recognition model was built.In order to overcome the shortcomings of the model,Adam gradient descent algorithm was selected to optimize the RBF neural network.The accuracy of this model is 87.43%,the recall rate is 86.60%,the accuracy is0.8807,the special effect is 0.8827,the false alarm rate is 0.1173,and the F1 score is0.8773.In Histograms of Oriented Gradients(HOG)algorithm,histograms of oriented gradients(HOG)algorithm are used to extract facial features from driving images.After that,the classification and judgment of face detection are realized by Support Vector Machine(SVM),and the classified face images are stored in the face image data set.A Convolutional Neural Network(CNN)recognition model is built,and the three-time convolution addition and one-time pool constitute a convolutional pool unit.At the same time,in order to prevent training and testing errors from becoming too large,Dropout is added to each fully connected layer to realize anger expression recognition.The accuracy of this model is 89.50%,the recall rate is 87.67%,the accuracy is 0.9100,the special effect is 0.8827,the false alarm rate is 0.0867,and the F1 score is 0.8930.A Fully Connected Neural Network(FCNN)anger emotion recognition model integrating vehicle driving data and facial expressions was established by integrating vehicle driving data and facial expressions through feature level splicing.The accuracy rate of this model was 91.32%.The recall rate was 90.67%,the accuracy was 0.9186,the special effect was 0.9197,the false alarm rate was 0.0803,and the F1 score was 0.9126.Then,on the basis of the two-source recognition model,a three-source fully connected neural network anger emotion recognition model considering language factors was established by integrating the language features and using feature level splicing.The accuracy of the model was 95.73%,recall rate was 97.73%,accuracy was 0.9397,special effect was 0.9373,false alarm rate was 0.0627.The F1 score is 0.9582.By comparing the advantages and disadvantages of the four models,the results show that the anger emotion recognition model of three-source fully connected neural network considering language factors has the best effect.These four drivers’ anger emotion recognition algorithms provide a more effective method for the study of drivers’ anger emotion recognition,and also provide some theoretical support for the improvement of vehicle safety assistance system and the realization of passenger-car sharing.