Research And Implementation Of Driver Drowsiness Detection System Based On Near-infrared Image

With the increasing vehic le ownership and the expanding highway coverage, traffic accidents happen more and more frequently, as one of the most important reason causing traffic accidents, fatigue driving has caught most people’s attention. How to accurately determine the driver whether or not in the state of fatigue in real-t ime, avoiding traffic accidents caused by fatigue driving has become a hot research topic among scholars and it has very signif icance to protect people’s property security and to improve the security of car driving.The main work of this paper includes three parts, face detecting from the images at night, fatigue parameters extraction and the determination of the fatigue. Firstly, in face detection, we thought fatigue driving happens more in the night, so we use the near-infrared image as input to detect fatigue driving. We detailedly studied and analyzed the face detection algor ithm based on Adaboost and the Adaboost cascade classifier learning principle, trained a cascade classifier using large amount of human face near-infrared image and implemented face detection. In terms of fatigue parameters extraction, we put forward an algorithm based on integral projection and image contour detection which can realize the real-time positioning of human eyes. Considering the variability driving conditions and the fluid dr iving posture, to improve the accuracy and robustness, we put forward a method to search eyes based on region growing when the eye locating is failed. Beside these, according to the detection acquirements, we implemented the blink detection based on interframe difference and yawn detection based on contour detection.Fusing the human eye open degree, eye blink and yawn, on the basis of a large number of simulat ions and exper iment summary, in this paper, we put forward three fatigue criterions to judge whether the driver is in fatigue state and accomplished the whole fatigue driving detecting system. The experimental results shows that the eye locating algorithm, the blink detecting and the yawn detecting method raised in paper have high accuracy and the fatigue detecting system can tell whether the driver is tired precisely, it satisfied the requirement of real-time, non-contact and high robustness and it has certain theory signif icance and practical use value.