Fatigue Driving Detection Based On Visual Feature Fusion

With the rapid development of the national economy,infrastructures such as roads,railways,and urban rails have gradually improved.More and more people choose to travel by car,train,and subway.These transportation modes bring great convenience and enjoyment to people's work and life.But with it comes a surge of road traffic accidents.According to statistical data,more than half of the world 's annual traffic accidents are caused by driver fatigue driving.Therefore,effective detection of the fatigue state of the driver during driving is of great practical significance for reducing or avoiding the occurrence of traffic accidents.The main work and innovations of this paper are as follows:(1)Aiming at the problem that the traditional fatigue detection targets are easy to be occluded and have individual differences,a fatigue driving detection method combining facial expression and head posture is proposed.This method studies the characteristics of fatigue expression and the method of using facial motion unit AU to express fatigue,and combines with the head posture to construct 8-dimensional joint fatigue parameters(AU?02,AU?09,AU?15,AU?17,AU?45,pose?yaw,pose?pitch pose?roll).Eight Statistics including maximum,minimum,average,range,median,variance,standard deviation and covariance of each one-dimensional parameter are calculated as the final fatigue feature vectors in the video frames.SVM classification model is established by combining with genetic algorithm,which optimized the penalty coefficient and kernel parameters of SVM to improve the detection effect.Finally,the video frames fatigue detection is carried out on the sliding window,and the precision rate is 91.42%,the recall rate is 91.5%,and the F1 value is 91.23%.(2)In view of the fact that the fatigue detection method is easily affected by light under visible light,a fatigue detection method based on infrared thermal image is proposed.The method is based on the specific expression of fatigue on the facial infrared thermal image,which caused by insufficient sleep in clinical medical research.The infrared thermal image data of the driver's face is obtained by establishing an infrared thermal image acquisition system platform,and the 15-dimensional texture features are obtained respectively by calculating the gray-gradient cooccurrence matrix for the eye area and the forehead area.The method uses three classifiers of KNN,SVM and Random Forest to establish the detection models.In the fatigue detection based on the infrared thermal image of the eye area,the best accuracy rate is 93.42%,the recall rate is 93.24%,and the F1 value is 92.86%.In the fatigue detection based on the infrared thermal image of the forehead area,the best accuracy rate is 93.66%,the recall rate is 93.12%,and the F1 value is 93.37%.(3)According to the unreliable accuracy based on single information source of fatigue detection,a fatigue detection method based on multi-source visual fusion information is proposed.The facial visual features(including facial expression and head posture)under visible light and the facial infrared thermogram(including eye area and forehead area)texture information are fused in the feature layer,and then the fatigue state is evaluated by visual and physiological signals.Aiming at the problem that high-dimensional features tend to be irrelevant and redundant which affects the classification results,a MI-FCBF feature selection algorithm is proposed.The algorithm combines mutual information and symmetric uncertainty,and removes irrelevant features and redundant features without affecting the classification effect,and obtains the best feature combination.The experimental results show that when the feature dimension is 30,the detection model based on Ada Boost achieves the best results,in which the precision rate is 94.83%,the recall rate is 94.9%,and the F1 value is 94.81%.