Driving Fatigue Detection Method Based On Deep Learning

With the increasing number of cars in China,the number of road traffic accidents remains high,which also leads to a large number of casualties and property losses.Road safety has become a hot topic of social concern and an urgent problem to be solved.Among them,driving fatigue is one of the main causes of road traffic accidents,and the proportion of major traffic accidents caused by fatigue driving is high,so the study of fatigue driving has very important significance.This study focuses on the design of a timely and reliable driving fatigue detection model to provide an accurate discriminant basis for real-time intervention of driving fatigue.Based on the NTHU-DDD data set published by Taiwan Tsinghua University,this paper uses deep learning technology to realize static fatigue detection of driving images and dynamic fatigue detection of driving videos respectively,which provides a basis for safety management of transportation units and accident identification of traffic management departments.In this paper,a driving fatigue detection model is designed based on static driving images.Firstly,on the basis of the multi-scene driving data of NTHU-DDD,the static frame image of driving video is extracted by Open CV and taken as the basic data set of the model.In order to improve the stability of the algorithm,image enhancement methods of turning over and adding noise were used to expand the data set,so that the fatigue detection model could be more robust under different scenes,different locations of image acquisition equipment and noise during data transmission.Then,the control variable method was used to calibrate the fatigue detection model based on Res Net50 network from the aspects of transfer learning,initial learning rate and batch＿size.After training,the accuracy of the calibrated model on the verification set can reach 99.3%,and the F1-score can reach 98.8%.Finally,the fatigue detection model of driving images is compared with the model built by Le Net,VGG16 and Google Net algorithm.Through analysis and comparison,it is found that the performance of the model built based on Res Net50 network is better than the other three models,and it is more suitable for fatigue detection of driving images.In this paper,a driving fatigue dynamic detection model based on driving video is designed.First,Open CV is used to cut the NTHU-DDD data set to create the basic data set of the model.In order to improve the stability of the algorithm,image enhancement methods of flipping and adding noise are used to expand the data set.Then,the control variable method is used to calibrate the fatigue detection model based on TSN network from two aspects: data preprocessing stage and model training stage.After training,the accuracy of the calibration model on the verification set can reach 95.9%,and the F1-score can reach 96.1%.At the same time,the fatigue detection model of driving video is compared with the model built by LSTM,I3 D and Slowfast algorithm.Through analysis and comparison,it is found that the performance of the model built based on TSN network is better than the other three models,and it is more suitable for fatigue detection of driving video.Finally,the driving video fatigue detection model is compared with other fatigue detection methods,and the results show that the model is more competent than other methods in the task of driving fatigue detection.The results show that the two kinds of driving fatigue detection methods proposed in this paper can effectively identify driver fatigue,and the research results can provide the core algorithm for the development of driving fatigue detection equipment,thus reducing the frequency of driver fatigue driving.