Research On Fatigue Driving Model Based On TensorFlow’s Framework Of Convolutional Neural Network

With the development of smart cars and the increasing number of domestic cars,the frequency of daily use of cars is also increasing,and people’s requirements on performance and safety coefficient are also increasing.At the same time,with the increase of car ownership,along with a series of traffic safety problems,and in many traffic accidents,driver fatigue driving has become one of the important reasons for the occurrence of traffic accidents.At present,many automobile companies and research institutions have taken driver fatigue test as an important subject and obtained some research results.Therefore,the inspection of insufficient fatigue of drivers can not only effectively reduce the probability of traffic accidents,but also be an important part of the intelligent development of automobiles in recent years.It has high scientific research and social and economic value,and can also promote the intelligent development of automobile industry.The main research object of this paper is eye state recognition as the research object.Through Google TensorFlow deep learning framework,a model suitable for convolutional neural network(CNN)is established to train and study the characteristics of eye fatigue state.At the beginning of this paper,the development history of neural network is studied,and some basic concepts and network structure of convolution upgrade network are introduced.Then introduces the concept and research theory of fatigue driving,as well as various algorithms.The data set of this paper adopts the open data set of NUAA University.The images of the acquired data set are normalized,and the data set is classified and saved,so as to improve the accuracy of the image in the recognition process,and the image is pre-processed with grayscale.Then,a neural network model is constructed by using Lenet5 convolutional neural network.According to the structure,the structure and parameters of the convolutional neural network model are adjusted accordingly.First,a group of comparison experiments are conducted from the size of the convolution kernel,and it is concluded that the size of the convolution kernel of this data set needs to be reduced compared with that of the Lenet5 network model.Also,according to the experimental results of the second group,it is concluded that the data set studied must reduce the number of convolution kernels to have a better effect.Optimized according to the result of the experiment,through the contrast test and according to the actual results is analyzed,and in combination with the practical situation of the specific data sets,and then the convolution layer number,convolution kernel size,number of convolution kernels after adjustment and the experimental parameters,such as,find the best neural network structure model of fatigue state detection,in the recognition accuracy and real-time performance to achieve the ideal effect,so that the driver fatigue detection ability,achieve the goal of reducing traffic accidents.Through the analysis of the final experimental results,the convolutional neural network model designed in this paper can well complete the training and research on the fatigue driving state of drivers,detect the fatigue driving state of drivers in the driving process and give early warning according to the detected results,thus reducing the probability of traffic accidents.To the country and society to reduce the tragedy and economic losses caused by traffic accidents.At the same time,it also shows that the research direction of this subject is relatively cutting-edge,and it can produce certain social and economic benefits.However,due to the limitations of research conditions and other reasons,there are many scenarios and hardware and software limitations,and further investigation and research will be needed in the future.