Study Of Vision-based Driver Emotion Recognition

The driver monitoring system is playing an increasingly important role in road traffic safety in recent years.Current driver monitoring systems mostly focus on monitoring drivers'fatigue and distraction,and pays little attention to the driver's emotional state that has a significant impact on driving behavior.Drivers should maintain a positive and stable emotional state during driving,paying attention to road conditions and vehicle conditions that may change at any time,due to the complex and changeable urban and rural road traffic.It's a difficult task for drivers to The difference of the driver's personal psychological quality makes it difficult to expect the driver to maintain a good driving emotion on his own,so it is necessary to monitor the driver's emotional state in real time during the driving process,and issue a warning reminder to prevent dangerous driving behavior when abnormal driving emotion is found.Therefore,this paper chooses the non-contact visual technology to monitor the driver's emotional state,summarized as follows:?1?Discussed the current research status of driver's emotion recognition and visual-based emotion recognition.By analyzing the relationship between micro-expressions and emotions,determined the thinking and research direction of this paper through micro-expressions to identify emotions.?2?Studied the related technology of vision-based driver's emotion recognition,and associated the category of emotion with the category of micro-expression.At the same time,divided vision-based driver's emotion recognition system into three steps:image preprocessing,feature extraction and emotion classification.Finally,the experimental method and performance measurement for evaluating the performance of the algorithm are discussed as the basis for subsequent research experiments.?3?An end-to-end deep learning method was proposed to recognize micro-expressions,and the pre-trained model in the field of face recognition is fine-tuned by the transfer learning method to solve the problem of data scarcity in training deep learning model.At the same time,the problem of class imbalance in micro-expression datasets is analyzed.Focus loss function is introduced as the objective function to reduce the impact of data imbalance.The validity of the algorithm model is verified by comparative experiments.?4?A method based on recurrent neural network extract micro-facial features in time domain and transferred model to extract spatial features is proposed to recognize micro-facial expressions in video image sequences.Time interpolation model is used to normalize the sequence.The obtained image frame sequence is used as the input of the recurrent neural network model.VGGFace-LSTM model is designed to extract spatial-temporal features of micro-expressions for classifying video emotion.By comparing the accuracy and ₁F-score of emotion classification,the validity and superiority of the proposed model are verified.