Abstract: This paper proposes a multi-level face detection scheme.It successfully combines the face detection and tracking approach，and is used in a fatigue judging system.Firstly，process the image by compensating illumination to reduce the impact of illumination.Then calculate the similarity of the skin in the first image to remove the background and make the frame binaryzation.Then，the knowledge-based face detection is employed to get the information of face organs.Local processing is obtained in the next image corresponding region based on the face information obtained from the first image.Motion vector can be built from the information of the former two images.Using the motion vector，estimate the position of eyes in the following images and check the exact position.Finally，PERCLOS is used to analyze the face detection results and determine the status of fatigue.Experimental results show that the speed of the proposed face detection scheme is 2.5 times faster than the conventional methods，and the scheme can do a real-time tracking on eyes and has high detection accuracy.

Key words: face detection, motion vector, fatigue evaluation, illumination compensation

摘要： 将人脸检测和跟踪技术有效结合并引入到疲劳判断系统设计中，提出了一种多级结构的检测方法。首先对图片光照补偿，减小复杂环境中光照不均带来的影响。然后对首张图像计算皮肤相似度并二值化来去除复杂背景，进而利用基于知识的人脸检测方法得到人脸器官信息；接着根据当前图像已获取的信息在下一图像帧中的相应区域进行局部处理；然后再根据两帧间的人眼位置关系建立运动向量，从而对后继帧中的人眼位置进行估计和校验。最后，利用PERCLOS方法来对人脸检测结果进行分析，判断是否处于疲劳状态。实验结果表明，所提出系统的检测方法处理速度是传统方法的2.5倍以上，可以对人眼进行实时跟踪，检测准确度高。

关键词: 人脸检测, 运动向量, 疲劳判断, 光线补偿