【作者】 李智；

【导师】 谢剑斌；

【作者基本信息】 国防科学技术大学 ， 电子与通信工程， 2010， 硕士

【摘要】 疲劳驾驶可能导致军机军车事故,影响军事后勤运输保障;疲劳也可能导致边防值勤疏忽大意而造成严重后果。能够及时检测出疲劳状态并发出警示信号的疲劳检测技术可以有效预防失误以及事故发生,在人脸表情自动识别方面具有广泛的应用前景。基于机器视觉的方法与其他检测方法相比,具有实时性、准确性、非接触性、适用性及经济性等方面的优势,成为当前研究的一个热点。针对全天候、全时段人脸检测的要求,本文提出一种以红外光照环境下的人脸视频图像为输入,基于AdaBoost的方法人脸区域检测算法。该方法首先利用人眼在红外光照条件下的红眼效应定位瞳孔;然后根据瞳孔定位剔除大部分非人脸区域;最后在候选区域内使用AdaBoost方法完成对人脸区域的检测定位。该方法简化了传统AdaBoost人脸检测的运算,大幅提高了人脸检测速度。针对以往疲劳检测算法普遍存在的受光照条件影响大、检测速度慢以及可靠性差的问题,本文提出一种基于AdaBoost的疲劳表情快速检测算法。该算法首先对输入的红外视频图像进行人脸检测定位,然后将眼睛、嘴巴分割出来,分别提取两个子图块的PCA特征,分别输入事先经过AdaBoost训练得到两个分类器进行判别,将两个分类器的输出进行或运算得到最终的检测结果。实验表明,本文算法正确率高,速度快,具有很好的泛化能力和较强的鲁棒性,能够满足实时应用要求。更多还原

【Abstract】 Fatigue driving may lead accident of battleplan and military vehicle to affect military logistic transportation. Fatigue may also lead negligence of the frontier defence duty to severe aftereffect. The fatigue detection technique which can detect the fatigue states and generates some warning alarms is considered as one of the most prospective applications of automatic facial expression recognition. Compared with the other detecting methods, the method based on machine-vision has many advantages such as: real-time response, veracity, non-contact, applicability, economy, which make it become the focus of current research.Aiming at round-the-clock face detection, this paper proposes a face detecting algorithm using AdaBoost method. At first, the pupils are located using the red-eye effect of eyes in infrared illumination. The pupils’location is used to eliminate a majority of background. Then using AdaBoost method detects and locates face in the remanent region. This method simplifies the operation of traditional AdaBoost face detecting algorithm, and boosts the speed of face detection, benefits subsequent procedure, and improves the efficiency of the entire algorithm.Aiming at the problem of the great influence of illumination and the low reliability, this paper proposes a fast fatigue detection algorithm based on AdaBoost. At first, the face is located in the infrared video image. Two sub-images which respectively contain mouth and eyes are segmented from face image. Then, the eyes’and mouth’s features which are extracted from the sub-images by PCA are classified by corresponding classifiers which are trained by AdaBoost earlier. Finally, the algorithm determines the state of the face is whether fatigued or not according to the output of the OR operation between the outputs of the classifiers. Proved by experiment, this algorithm has not only high correct rate and fast speed but also a powerful ability to generally use and kind of robustness. And the response time of this algorithm could satisfy the real-time requirement.更多还原