Research On Driver’s Emotion Recognition Algorithm Based On Machine Vision In Complex Illumination Environment

The driving behavior of the driver in a negative emotional state is one of the important causes of traffic accidents,especially the "Road Rage" phenomenon caused by negative emotions,which seriously affects the driver’s safety.Therefore,it is of great significance to do research on the theory and method of driver emotion recognition,monitor the driver’s emotional state in real-time,and actively intervene the negative emotions of the driver when they appear,to guarantee the driver’s personal safety and road traffic safety.At present,the research on driver’s emotion recognition mainly stays in the laboratory simulation method,the scale method,and the physiological signal detection method.These research methods cannot realize a real-time driver’s emotion recognition without affecting the driver’s normal driving.Aiming at this problem,this paper proposes a real-time dynamic emotion recognition method based on machine vision.The method first uses the theory of machine vision to identify the facial expressions of the driver in the driving state,then studies the emotional state shown by the expression change,and obtains continuous emotional changes on the time axis.This paper creatively proposes the concept of driver emotional index(hereinafter referred to as "emotional index")to measure whether the driver is suitable for driving at this time.The introduction of the emotional index makes it possible to study the emotional changes shown by the continuous changes in expression.In response to the driver’s emotion recognition problem,this paper has done three aspects.(1)Through the analysis of the current classical image enhancement algorithm,this paper proposes an AAQR(Adaptive Attenuation Quantification Retinex)illumination enhancement algorithm based on the Retinex theory.By predicting the image noise range,the algorithm autonomously attenuates the useless information in the image,re-quantizes the effective features of the image,and enhances the feature information in the image,which can ideally solve the influence of ambient illumination on the face image.At the same time,the AAQR algorithm solves the halo phenomenon of the image enhancement algorithm based on Retinex theory to some extent.(2)To improve the accuracy of expression recognition and to retain timing information for recognizing expressions,this paper proposes a multi-structure variable parameter expression recognition modelMVCNN(Multi-structural Variable-parameter CNN)based on the CNN model.The MVCNN model improves the model over-fitting and gradient dispersion problems caused by the high number of layers in the CNN model through the "simple and complex" neural network structure,and solves the problem that the traditional CNN expression recognition model does not retain the recognition of expression timing by the way of sorting the recognized expressions.In addition,a new image preprocessing method,image block preprocessing,was designed.The preprocessing method enhances the direct connection between the local and local parts of the image and helps to improve the robustness and generalization of the model.(3)A system of emotion recognition based on facial expression timing information is proposed.This paper firstly discusses the positive and negative emotions of the driver by analyzing the expression of individual emotions and the driving situation that the corresponding driver may have.Secondly,by finding the basic correspondence between the driver’s expression and emotion,the correlation between the seven facial expressions with the negative and positive emotions is quantitatively processed.Moreover,the emotional index is used to express the emotional state of the driver.Then,with the emotional index the conversion relationship between facial expressions is quantitatively analyzed for the operation behavior of the driver’s.And the facial expression conversion table is established to realize the continuous expression of discrete facial expressions.Finally,the driver’s emotional index is obtained by looking up the conversion table and the emotion recognition algorithm to recognize the driver’s emotion.The goal of this paper is to solve the problem of driver’s emotion recognition in complex illumination environment,to explore the correspondence between driver’s facial expression and emotion,to evaluate the emotional state of the driver while driving,and to provide the theoretical and practical basis for a practical application.