Research And Implementation Of Driving Fatigue Detection Based On Facial Feature Tracking

With the increase of vehicles and the frequent occurrence of traffic accidents,in order to reduce the incidence of accidents,active prevention and control should be taken.As one of the important culprits of traffic accidents,fatigue driving may cause traffic accidents at anytime and anywhere in the world.For this reason,many countries in the world have carried out research on the prevention of fatigue driving.Among many detection methods,fatigue detection based on computer vision is favored by many researchers because of its non-contact,good robustness,strong real-time performance and high accuracy.The key and difficulty of fatigue detection based on computer vision lies in the accurate and real-time detection and tracking of facial features.In this paper,the ellipse skin model is used to roughly locate the face,and Adaboost is used to precisely locate the face.Experiments show that this method saves more than half of the time by directly detecting the face area in the image.The improved Mean Shift algorithm and Kalman filter are used to track the face region.Under the conditions of light interference,fast motion and static occlusion,the improved algorithm can accurately track the face position.In the test video with a resolution of 480*640,the average time of tracking one frame is 22 ms.The human eye and the mouth are detected and positioned in the tracked face area.The eye state is judged by the aspect ratio and the black and white pixel ratio,and the mouth state is judged by the aspect ratio,thereby judging whether there is blinking or yawning event.The blink frequency,the number of yawnings and the number of consecutive closed frames of the eyes and the PERCLOS value were recorded within 30 seconds.The driver's fatigue status is judged by the above four indicators.In the 5 self-made videos,the methods in this paper can accurately determine the phenomenon that occurs in the video,and judge whether the alarm is needed according to the phenomenon.In the face tracking experiment and fatigue judgment experiment,this paper simulates the possible conditions in the driving process.In the test video,the algorithm has good performance.For a video with a resolution of 480*640,the average processing time of a frame is 38.9ms,which meets the real-time requirements.