Research On Driver Fatigue Detection Method Based On Vision

With the continuous increase of car ownership,traffic accidents have become a social problem threatening personal and property safety.As the pace of life continues to increase,the number of traffic accidents caused by fatigue and drowsiness is increasing year by year.Fatigue detection can reduce the occurrence of traffic accidents.This paper uses computer vision technology to analyze driver’s facial image and study a real-time and accurate fatigue detection method.The algorithm consists of four main steps,namely,face detection,face alignment,eye and mouth area location and state recognition,and multi-parameter fusion fatigue detection.The main contributions of this thesis are as follows:(1)In order to improve the speed of Adaboost face detection algorithm,this paper proposes a search method based on anticipation of face location.Firstly,the face is detected on the whole video image,and then the largest face image area detected in the previous frame is expanded as the detection area of the next frame.This method removes a large number of background areas and improves the speed of face detection.(2)Facial landmarks are quickly and accurately detected on the driver’s face image.Face images marked with 5 landmarks is used to train facial landmarks detection model based on ensemble of regression trees(ERT)and multiple convolution neural network(CNN)facial landmarks detection networks with different structures are designed.The effects of different parts of CNN and their influences on detection results are analyzed.The facial landmarks detection model suitable for this paper is selected through experiments.(3)An eye and mouth state recognition method based on BLS(broad learning system)is proposed.First,the eyes and mouth areas are extracted according to the landmarks of the face,and then the eyes and mouth images are stretched into a onedimensional vector as the input of the broad learning network to identify the state of the eyes and mouth.The experimental results show that the algorithm is simple to train and has high recognition accuracy.(4)A fatigue detection method based on broad learning system is proposed.First,eyes state,mouth state and head state are judged,and then the time sequence of state of eyes,mouth and head are combined as a parameter for measuring fatigue,finally the broad learning system is used to judge the state of the driver.(5)A driver fatigue detection simulation system based on PC and a driver fatigue detection system based on Raspberry Pi are realized.The experimental results show that the fatigue detection algorithm proposed in this paper has high real-time and accuracy,which can meet the needs of practical applications.