Research On Driver Emotion Recognition Based On Physiological Signals

With the arrival of the intelligent era,the functions of automobile have gradually diversified,but its safety performance has always been of great concern to people.At present,the main methods to improve the safety performance of automobile are real-time location,real-time monitoring,real-time diagnosis and real-time road condition prediction,which are preventions of external accidents.The lack of evaluation of driver’s own state is in fact a problem.The existing efforts to evaluate driver’s own state are mainly fatigue detection and stress detection.The fact that driver’s ability to perceive,decide and react to the external environment is highly influenced by his or her emotions,makes it necessary to take the driver’s emotional factors into account to improve the safety performance of automobile.However,the limited research on driver’s emotions so far is either more focused on a single emotion of anger,or limited to the impact of emotions on driving behavior,or designing a driver’s emotion detection system based on general emotion recognition algorithms without experiments designed to collect emotion data while the driver is driving.which differs from the actual situation and the conclusions are bound to be biased.In this article,we broadened the types of emotions to be studied which include calm,happiness,sadness and fear,and conducted research that focus on driver’s emotion recognition.A driving simulation experiment was designed to collect the actual data of the driver’s emotions during driving condition.Three common peripheral physiological signals including electrocardiogram(ECG),respiration(RSP),blood volume pulse(BVP)were chosen to recognize driver’s emotions due to the consideration of recognition effect,experimental cost and research workload.In addition,we try to conduct an exploratory study on the emotion regulation of drivers’ fear as an example since the emotion regulation after emotion recognition has not been paid much attention by researchers.At the beginning,we conducted research on signal pre-processing,feature extraction and dimensionality reduction techniques which formed a segment for physiological signal processing based on some typical physiological signal samples referring to the experience of previous studies.Subsequently,we selected the DEAP dataset which is commonly used and highly recognized and used the BVP and RSP signals in it to construct the base model for emotion recognition based on the support vector machine algorithm through the physiological signal processing which is previously mentioned.And then optimize the two hyperparameters of the model,box constraint and kernel size,with Bayesian optimization algorithm.Finally,the results show that the base model achieves 9.38% and 9.93% increases in average accuracy of valence and arousal respectively.After that,we designed a driving simulation experiment.A platform was firstly built for physiological signal acquisition.And then we collected ECG,RSP,BVP signals based on the platform and acquired subjective evaluation data of drivers who were induced with four emotions,and introduced regulation measures for drivers’ fear to obtain data required for the exploratory study on drivers’ emotions regulation during the experiment.The physiological data are processed by the physiological signal processing before to obtain the effective feature matrix,which was used to expand the input of model.In this way,the ECG-RSP-BVP model for driver’s emotion recognition could be formed based on the architecture of base model constructed in the previous section.Some relevant metrics were selected to evaluate the recognition performance of the model.Then six classification tasks of pairwise combinations of four emotions were implemented based on the model.The results show that the model generalizes well on the test sets of six classification tasks.All the average accuracy is more than 84%,among which the average accuracy of the classification task between calm and happiness is close to 90%.At the end of the study,we designed four regulation schemes to regulate drivers’ fear emotions based on the theories related to emotion regulation.It is shown that all the four schemes could achieve a certain effect on emotion regulation through the analysis of subjective evaluation data and physiological data.