Research On Fatigue Driving Detection Method Based On Facial Feature Extraction

Fatigue driving has caused a lot of traffic accidents.Further study of fatigue driving and real-time monitoring of the driver’s state are of great significance for improving driving safety.At present,as one of the fatigue driving detection methods,the fatigue driving detection method based on facial features has the advantages of real-time,non-intrusive,low cost,and high accuracy,and has become popular in the research of fatigue driving detection methods.However,traditional fatigue driving detection methods cannot identify the fatigue behaviors based on time-series features effectively,such as yawning and drowsiness.At the same time,there are interference factors such as the changing illumination condition,extreme head pose and face occlusion that severely affect the accuracy and computational efficiency of the fatigue driving detection method based on facial feature.In view of the above shortcomings,this paper proposes a fatigue driving detection method based on facial feature extraction and feature discrimination of support vector machine(SVM)combining traditional machine learning and deep learning methods,and develops a fatigue driving detection system based on facial feature extraction.The main works are as follows:(1)To improve the accuracy of face detection in fatigue driving detection,this paper proposes a face detection method based on histogram of oriented gradient(HOG)features and Support Vector Machine(SVM).Firstly,the self-quotient image is calculated on the original image to eliminate the influence of the changing illuminations.Secondly,compared with the Haar features and the local binary pattern(LBP)features,the HOG feature has the advantages of illumination invariance and geometric invariance,further improving the accuracy of face detection under harsh conditions.Experimental results show that the proposed method improves the accuracy of face detection effectively.(2)To improve the accuracy and speed of facial landmark detection,this paper proposes a facial landmark detection and tracking method based on cascading constraint local model.A cascaded landmark point distribution model is trained to extract the eye and mouth regions quickly and accurately that are related to fatigue behaviors.To improve the accuracy of facial landmark in important regions,the cascaded constrained local model adjusts the average facial landmarks roughly using the full-face landmark point distribution model,and then fine-tunes the landmarks of the eye and mouth using the eye-mouth landmark point distribution model.To improve the detection speed of the landmarks,this paper sets up a face evaluation model to improve the accuracy of the initial position of the facial landmarks,thereby reducing the number of iterative searches for the facial landmarks.Experimental results show that the proposed method improves the detection accuracy and speed of facial landmarks effectively.(3)To overcome the individual differences of drivers and improve the accuracy of yawning detection,a yawning detection method based on residual network(Resnet)and long short-term memory(LSTM)network and an eye state recognition method based on convolutional neural network are proposed in this paper.The yawning detection method based on Resnet and LSTM extracts the spatial information of the image through the Resnet,and extracts the time-series information between frames through the LSTM.Experimental results prove that the fusion of spatial-temporal information can effectively distinguish the driver’s actions such as yawning,singing and speaking,improving the accuracy of yawning detection.The eye state recognition method based on convolutional neural network overcomes the individual differences in the size of the driver’s eyes.The proposed method automatically extracts the spatial features of the eyes through the convolutional neural network,avoiding errors caused by the traditional manual calculation of the eye-opening size.Experiments show that the proposed method improves the accuracy of eye state recognition effectively.(4)To improve the accuracy of fatigue driving detection,a fatigue driving detection method based on facial feature fusion is proposed in this paper.Firstly,the facial features of the drivers in the state of awake and fatigue are analyzed to construct a feature vector space which is related to fatigue behaviors such as yawning and drowsiness.Secondly,a classification model based on SVM in this paper is trained to discriminate the fatigue behavior according to the extracted feature vector,thereby reducing the risk of overfitting in the discrimination process,and improving the robustness and accuracy of fatigue driving detection.Experimental results show that the proposed method recognizes the driving state of the driver effectively.(5)To verify the effectiveness of the fatigue driving detection method mentioned above,a fatigue driving detection system based on facial feature extraction is developed independently in this paper.The basic requirements of fatigue driving detection are met by testing the system.The experimental results show that the methods proposed in this paper improve the accuracy of fatigue driving detection and recognize the state of the driver in real situation effectively.