Research On Driver’s Road Rage Emotion Recognition Based On Video Images And Deep Learning

"Road rage" is a disorder of intermittent rage burst,which means that the driver is driven by anger with the influence of the external environment.This emotion can easily cause the driver to have aggressive driving behavior,which ultimately affects traffic safety and even causes traffic accidents.Therefore,it is of great practical significance to identify drivers’ road rage emotions and reduce the impact of road rage on traffic safety.The methods for identifying road rage emotions of drivers can be divided into two categories.One is a subjective survey method based on interviews and questionnaires.The driving rage scale is used to assess whether drivers have road rage emotions.This method is simple and effective,but not real-time.The other method is to collect data such as facial images or physiological signals of drivers,extract features,build models,and objectively analyze driver’s road rage emotional state in real time,which is the mainstream of the current research direction.Combined with previous research,this paper applies the deep convolutional neural network method with excellent performance in the field of image recognition to road rage emotion recognition,giving a driver road rage emotion recognition method based on video images and deep learning.The main research contents of this article are as follows:(1)Face area detection based on HOG image pyramid features.Based on the images with face tags and non-face tags,extract the HOG image pyramid features and the SVM method is used to train the model.The non-maximum suppression method processes the overlapping detection frames and finally obtain the driver face detection model.After the model obtains and crops the facial area image,the image size is normalized,which is modified to 3*299*299 pixels for further processing by the deep convolutional neural network.(2)Using the transfer learning method and InceptionV3 model,the driver’s road rage emotion recognition model based on facial images is constructed.The expression image sequences in the CK+database are screened and face detected,and samples of angry faces were enhanced.Finally the angry and non-angered face image data set is constructed,with a total of 510 angry expression pictures and 542 non-anger expression pictures.Based on the InceptionV3 deep convolutional network neural network model pre-trained by the ImageNet database,transfer learning training is performed to obtain a road rage emotion recognition model that can determine whether a single face image is in an angry state.After training for 300 steps,the average recognition rate of the model on the test set is about 97.7%.(3)The method of road rage emotion recognition based on video image sequence is studied.In a sequence of driver facial images obtained through a camera,if three or more consecutive pictures are recognized as angry expressions,or the proportion of angry expression images exceeds 40%,the driver is judged to be in a road rage state within the current time window,otherwise in a non-rage state.The video image sequence is updated and obtained by the sliding window method.Each time 50%of the images in the image sequence are updated,which can enhance the accuracy and stability of the model.The research results in this paper can provide theoretical support for advanced driving assistance system.When it is recognized that the driver is in a road rage emotional state,soothing music or preset audio can be played to appease the driver’s emotions.When the situation is urgent,emergency measures can be taken to warn the driver,or the automatic assisted driving system can take over the vehicle driving.