| The fluctuation of emotional state will affect the reception and judgment of traffic information,thus affecting the accuracy of operation.Therefore,traffic accidents are often caused by emotional problems in the process of driving.However,at present,there are few researches on the influence of different emotions on driving strategies,and the automatic detection and warning of emotions mainly focus on "road rage".It is of great theoretical significance and application value to study the driver’s emotional experience and conduct real-time identification and timely warning for avoiding many traffic problems caused by emotional driving and realizing accurate protection of people and vehicles.Facial expression and speech information are the main channels for drivers to express emotions.Based on data analysis,this paper studies the problems of facial expression recognition and speech emotion recognition,and proposes a driving emotion recognition method based on facial expression and speech information.Specific research work is as follows:(1)Research on emotion and driving behavior.Based on domestic and foreign research,the driver’s emotional expression scale was designed and tested to obtain drivers’ various emotional experiences under the emotional driving situation,and the importance of exploring anger,fear and happiness was determined.The behavioral characteristics of drivers with different emotional states were analyzed from the aspects of operating frequency and intensity.Starting from the consequences of emotional driving,the influence of emotion on the driver’s physiology and psychology is studied.Finally,the influence of emotional driving on traffic safety is discussed,and it is found that the diversity of emotional expression will affect the behavioral operation and physiological state of drivers,leading to various forms and serious accidents.(2)Face detection and expression recognition.Facial expression directly reflects the emotional state,is the main channel of emotional expression.Firstly,a face detection model based on HOG pyramid feature and support vector machine is established to overcome the interference of illumination and noise,and reduce the sensitivity of local deformation.Then,the Inception V3 model is used as the basic framework,and the transfer learning training mechanism is used to establish the emotion recognition model based on facial expression,which has a better recognition effect on the enhanced expression sequence of samples.(3)Speech emotion recognition.Phonological features also have high value in representing emotions.Firstly,the speech signal is preprocessed by endpoint detection,pre-weighting,frame division and window addition to obtain even and smooth signals.To realize speech emotion recognition,the effective extraction of speech features is extremely important.Therefore,on the basis of in-depth analysis of the correlation between the acoustic features of speech and drivers’ different emotional states,the optimal extraction and calculation of its statistics are carried out to achieve the extraction of global emotional features.Finally,emotion recognition based on speech information is realized based on sample entropy and five basic acoustic features.(4)Bimodal emotion recognition.Considering the diversity of drivers’ emotional expression,a two-modal emotion recognition method based on facial expression and voice information fusion was proposed to address the deficiency of single modal information.Facial expression features and voice signal features were extracted respectively for decision-level fusion,and voice emotion was used as an aid for expression recognition,which made up for the deficiency of driver emotion recognition only relying on single modal information,and effectively improved the accuracy of emotion recognition. |
