Research On Night Fatigue Driving Detection Method Based On Facial Feature Tracking And Recognition

Traffic accidents have lasted for many years,causing huge losses in social production and people's lives.Fatigue driving behavior is one of the important causes of this situation.The causes of human fatigue are complex,the time of generation is uncertain,and it is difficult to circumvent the behavior through policies and regulations.Therefore,it is of great significance to find a fatigue detection method based on active safety technology.At present,active safety technology is a cutting-edge research content in intelligent transportation.In this thesis,the active safety technology for fatigue driving detection is studied.The machine vision perception method is used to analyze the facial features of the driver during nighttime when people get drowsy easily.The main research contents are as follows:(1)Improved Viola-Jones face recognition method based on Gentle AdaBoost and Normalized Pixel Difference(NPD)features is proposed.On the basis of Viola-Jones face recognition framework combined with the requirements of multi-pose face recognition,the AdaBoost algorithm and Haar-like rectangle features are replaced.In the improved method,the generalization ability of Gentle-AdaBoost is better,and the NPD is more suitable for describing multi-pose face.They can improve the robustness of face detection.Then the near-infrared face images are used to train and verify the classifier.Experimental results show that the improved face recognition method gets higher detection rate,lower false detection rate and lower missed detection rate.(2)Eye positioning method based on distribution feature and texture feature is given,and eye tracking method based on the fusion of scale adaptive Mean-Shift algorithm and Tracking-Learning-Detection algorithm(Mean-Shift TLD)is proposed.According to the distribution feature of the eye in the face,the human eye is located roughly by image morphology processing.Then an accurate eye positioning method based on eyebrow,eye and skin texture features is proposed.The tracking module,the learning module and the detection module of the original TLD algorithm are analyzed,and it is found that the optical flow algorithm of the original tracking module has three characteristics needs to be improved: large amount of computation,weak resistance to deformation,and weak resistance to light changes.These problems are improved by combining the scale-adaptive Mean-Shift algorithm.The contrast experiment is designed,and through tracking the eye in the near-infrared nighttime video,this algorithm has achieved good tracking effect.(3)Fatigue detection method based on eye and mouth characteristics is studied.The area and contour of mouth are extracted by analyzing the facial distribution of the mouth and using image morphology processing.Then a method of yawn detection based on circularity is proposed to discriminate the fatigue.A calculation method of eyelid opening based on ellipse fitting is proposed.Through counting the changes of the eyelid opening degree,the PERCLOS fatigue discrimination method is used to identify the fatigue.(4)The prototype system is designed,implemented and tested.C++ language with open source machine vision library OpenCV is used to develop this nighttime driver fatigue recognition prototype system.The collected near-infrared video is used to test the system.The feasibility,real-time and nighttime applicability of it are verified.The thesis is supported by the third subject of the "National Road Traffic Safety Technology Action Plan(Phase II)"(No.2014BAG01B03)— "Development of key driver traffic behavior analysis technology and system",which is jointly developed by the Ministry of Science and Technology,the Ministry of Public Security,and the Ministry of Transport.The driver's driving behavior spectrum is a new concept put forward in the field of traffic safety.Driver's fatigue driving behavior is one of the characteristic indexes of driver's driving behavior spectrum.The research results provide a theoretical method for quantitative analysis of this characteristic index of driver's driving behavior spectrum.