Research On Vision-Based Driver Fatigue Detection

With the development of social economic, traffic accidents caused by driver fatigue have become one of the main factors. Therefore, how to monitor and prevent fatigue driving effectively is a important practical significance to reduce traffic accidents and mortality.In reference to a large number of domestic and foreign literature, compare various methods of fatigue detection, study the method of vision-based driver fatigue detection deeply which is non-contact,real-time including face detection,eye location,eye states detection,eye tracking and fatigue analysis.Firstly, use color face algorithm which is real-time to detect face and propose a new binary method-accumulated histogram threshold method which reduce sensitivity to light. Secondly, color based iris detection is proposed, if has not iris, then use convolution template to locate eyes with accuracy greatly improved.Thirdly, propose a coarse-to-fine eye states detection method, whether there are iris to determine eye states, if not, then recognize eye states according to eyelid curvature which increases detection accuracy.Fourthly, Kalman filtering and mean shift algorithm based eye tracking method which can track close eyes and change size of eyes is proposed. Finally, methods of PERCLOS and blink frequency combination is proposed to improve the reliability of alarm.Experimental results show that these methods can used to recognize eye state and detect fatigue in the condition of head tilt, light changes, expression changes with high accuracy and good real-time performance. Eyes locating accuracy rate is 94.67 percent, state recognition accuracy rate is 98%, the average processing time for each frame is about 50 ms.