Research And Implementation Of Fatigue Driving Detection System Based On Eye Features

In recent years,with the increasing number of automobiles,the incidence of traffic accidents is also increasing.As one of the main factors inducing traffic accident,it is of great practical significance to study how to detect fatigue driving quickly and accurately.Because of the best correlation between eye features and fatigue,eye positioning and state of discriminant are the key steps in the fatigue detection methods.However,the traditional methods are highly susceptible to changes in illumination,occlusion,posture and other factors.In view of the above problems,this paper combines eye features with in-depth learning methods to conduct fatigue driving detection,which mainly includes face detection,eye location,eye state recognition and fatigue determination,etc.Firstly,the driver’s face is detected by MTCNN face detection algorithm based on in-depth learning;Secondly,an eye-positioning network model based on multi-task constraint learning is proposed for the deficiency of human eye positioning in MTCNN;and the LeNet-5 network model is modified to form the driver’s eye state recognition network,After obtaining the eye state,the driver’s fatigue state is determined by calculating PERCLOS value and blink frequency.Finally,PyQt5 and Tensorflow are used to implement a fatigue driving detection system based on eye features.In this paper,the relevant experiments are carried out on each module of the fatigue driving detection system.The experimental results show that the MTCNN algorithm has better accuracy and real-time performance than other methods on the FDDB and WIDER FACE face detection benchmarks.it is also highly robust to effects such as illumination,complex background and so on.In terms of human eye positioning,Compared with the traditional cascade CNN method,the proposed eye location model based on Multi-task constrained learning not only achieves better positioning effect under the influence of wearing glasses and head posture changes,but also has better real-time performance.In terms of eye state recognition,the average recognition accuracy and AUC of the eye state recognition network model can reach 97.28% and 99.49% respectively.Compared with the simple use of PERCLOS to determine fatigue,the more accurate fatigue state determination results can be obtained by combining PERCLOS and blink frequency.The accuracy of fatigue state detection under wearing glasses and without glasses is 95.23% and 96.75%,respectively.Moreover,the fatigue driving detection system implemented in this paper can basically meet the requirements of real-time.