Study On Multimodal Recognition Method Of Driver’s Anger Emotion And Mechanism Of Driving Risk Under Anger Emotion

In the modern road traffic system,the driver plays the most important role in the closed loop of the traffic system of the road,so the driver’s own state and driving performance will directly affect the safety of himself and other participants.The driver’s state is determined by many factors such as emotion,fatigue state and task load,among which the driver’s emotion is particularly important.Studies have shown that "road rage" caused by anger has a significant negative impact on driving risk perception,evaluation and decision-making,and decision-making execution,and is prone to driving safety risks.Road rage has also become an important cause of road traffic accidents.Therefore,anger emotion research has become a hot research of driver emotion recognition.However,in the existing research,at the theoretical research level,more attention is paid to the correlation between different emotion models and physiological data;at the experimental research level,there is a lack of research on driver emotions in the driving environment;at the emotion recognition level,mainly based on physiological data and EEG The data is the main research object,and there is a lack of research on driving behavior recognition and multi-modal combined emotion recognition in driving or simulated driving environment.Therefore,in view of the lack of existing research,this paper establishes a multimodal experimental measurement and recognition method of driver anger emotion based on the trinity of EEG signals,physiological signals and driving behavior information.While establishing a driver anger emotion data set,it uses attention-based The LSTM model of mechanism models and recognizes anger and neutral emotions.This paper is composed of four parts: emotion induction material screening experiment,simulated driving environment emotion induction and data collection experiment,single-mode LSTM emotion recognition modeling,and attention-based LSTM emotion recognition modeling at the feature layer and decision layer.In the emotion induction screening experiment,subjective evaluation was performed on the DES scale of the discrete emotion model and the SAM scale of the dimensional emotion model of the 42 groups of 42 videos in 7 groups,and the best audiovisual induction materials corresponding to the 7 emotions were found.In the simulated driving emotion induction experiment,an experimental method with simulated driver as the main body,which integrates audio and video collection,EEG data collection,driving behavior data collection and multi-channel physiological data collection,was established.It provides a complete,reasonable and effective measurement method for driving emotion research.At the same time,in the experimental data,mining the driver’s brain cognitive mechanism and driving safety impact model under anger,further confirming the negative impact of anger on driving safety at the physiological and behavior levels.Before recognizing the driver’s anger and neutral emotion models,data preprocessing and feature extraction were performed on 32 channels of EEG,18 channels of simulated driver and 3 channels of multi-channel physiology,respectively,to form a driving anger emotion dataset for 41 people Since the multi-channel physiology instrument had insufficient interface data forwarding digits in the first experiment and the test could not be repeated during the epidemic,the method of multi-channel physiology instrument data retention was deleted.Therefore,after each mode of single-mode LSTM emotion recognition,the correct rate of EEG signal anger emotion recognition is 73.7%,F1-score is 69.6%;the correct rate of driving behavior data anger emotion recognition is 65.8%,F1-score is 65.6%.Using the attention mechanism,the multi-modal LSTM is separately modeled and recognized at the feature layer and the decision layer,where the recognition rate at the feature layer is 74.2% and the F1-score is 70.0%;the decision layer recognition correct rate is 76.3% and F1-score73.2%.In the recognition of emotion categories,compared with the recognition methods of single-modal and multi-modal feature layers and multi-modal decision layers,the fusion mode of multi-modal feature layers shows relative superiority.Finally,based on the methods of driver emotion and anger emotion recognition,the academic and engineering significance is discussed.For the existing on-board driving ability assistance system,it is necessary to establish an emotional caring for the car with adaptive ability that takes human as the core System to realize the intelligence and intelligence of cars,let smart cars evolve from smart perception to smart cognition,and make driving safer.