Research On Driver State Analysis Method Based On Facial Expression

It is crucial for drivers to drive in a fit state for the sake of traffic safety when it comes to driving safety.Usually,the way to test abnormal driving can be divided into contact detection and non-contact detection,in which contact detection judges the drivers' driving state with various devices to check human physical signals while noncontact detection mainly inputs the drivers' facial video into the internet via web cams.The positions of facial key points and whether the drivers yawn,close eyes or not are the common method of image detection to judge the drivers are tired or not.Therefore,in this paper,the author has put forth a method of concatenated convolutional neural networks with the discovery of 68 key points in drivers' faces.However,this method has deficiency in testing efficiency and fails to take full advantages of men's rich facial expressions.Therefore,aiming at the above problems,this paper has brought up an analytical approach of the drivers' state detection based on the facial expressions.Facial expressions serve as one of the most important and direct carriers in expressing human emotions.According to the analysis of facial expression,much effective information can be inferred.At present,most excellent methods of facial expression recognition are on the basis of the convolutional neural network.In order to improve the accuracy of facial expression recognition,incremental convolution neural network will be applied in a frequent manner.However,only relied on increasing network depth will lead to enormous arguments which cannot achieve real-time requirement in the ordinary hardware configuration.In this paper,the author,targeted at the difficulty in instantaneity,puts up a deep neural network architecture.In accordance of concatenated network connections,it can not only decrease the arguments under the situation of depth increase,but also guarantee the testing accuracy.In this paper,the network contains three convolutional layers,one pooling layer and four improved inception layers(the improved method is to deform the size of inner convolution kernel).There is a pooling layer after each inception layer,finally outputing the result through softmax.The neural network can be applied to different hardware scenarios by adding or subtracting inception,which expands the application range.This paper has made contrast between today's popular models in the aspect of facial expression recognition.In the process of comparison,some modifications have been made to the models adapting to the hardware environment of this paper.The facial expression databases used are the mainstream facial expression databases.According to the experiment result,it has been proved that the structure based on the inception can ensure the accuracy rate of facial expression recognition together with simplifying the model within utmost degree.In the final,in this paper,an emi-physical simulation platform has been established;furthermore,experiments on the drivers' facial key points as well as their facial expression recognition have been carried out on the basis of this platform.In the last,in this paper,the author has given an eloquent proof that the analytical method of drivers' state funded on facial expressions can be well applicable for drivers' state detection.