基于面部特征分析的疲劳驾驶检测算法研究

发布时间：2017-12-31 08:02

本文关键词：基于面部特征分析的疲劳驾驶检测算法研究　出处：《中北大学》2017年硕士论文　论文类型：学位论文

【摘要】：在现有基础上人们对生活质量的要求愈来愈高,为了达到目标便不可避免地需要长期工作而导致劳累、困乏,在静态情况下发生困乏只需要休息就可以了,而当驾驶员在疲劳状态下继续驾驶车辆,则可能发生不可挽救的灾祸。疲劳不能够消除,但可以提醒,只要能在疲劳之初对驾驶员提出一定的示警,即可大概率避免交通事故的发生。为了实现此功能,就必须对驾驶员的状态进行实时的监控,而非接触性的状态监控方式又是诸多方法中最为合适的一种。本文针对相关技术进行了研究创新,对驾驶员面部特征进行了提取分析,具体工作如下:(1)对四种肤色提取方法进行实验,最终选定两种表现效果较好的肤色检测进行权值组合,实验表明组合后的肤色检测可以在相同时间复杂度下获得更优良的准确度,是单一肤色检测产生像素偏差的一半以下。(2)提出一种新的驾驶员疲劳检测方法,通过常识及实验分析发现,正常驾驶与疲劳驾驶的行为习惯会有所不同,即在正常驾驶时驾驶员会在一定时间内观察各个后视镜,而疲劳驾驶状态下该行为会被弱化甚至遗忘,通过人脸定位得到人脸运动曲线图来判断驾驶员是否处于疲劳状态。(3)学习了传统的人脸特征点定位算法,在SDM人脸特征点检测的基础上对眼睛特征点进行跟踪,提出了基于人眼拓扑结构的人眼特征点跟踪,能够快速得到人眼特征点位置,在此基础上提出了基于PERCLOS判定准则的人眼宽高比值分析和人眼区域黑色像素面积分析的疲劳驾驶检测方法,实验证明本文算法具有很好的准确性。
[Abstract]:On the basis of the existing quality of life that people demand more and more high, in order to achieve the goal will inevitably require long-term work and lead to tired, sleepy, sleepy occurred in the static case just need to rest it, and when the drivers in fatigue state continues to drive the vehicle, there may be irreparable disaster. No fatigue to eliminate, but can remind, as long as in the beginning of the driver fatigue some warning, you can probably rate to avoid traffic accidents. In order to achieve this function, it must be real-time monitoring of the status of the driver, rather than state monitoring contact and many methods one of the most suitable this paper studies the innovation. The related technology, the driver's facial features are extracted and analyzed, the specific work is as follows: (1) four kinds of color extraction method of experiment, finally selected two The weight combination performs better skin color detection, skin detection experiments show that the combination can obtain better accuracy at the same time complexity, is half the single color detection generates pixel deviation below. (2) proposed a new method for driver fatigue detection, through the analysis of knowledge and experiments, normal driving with the fatigue driving behavior will be different, that in normal driving the driver will observe all mirrors in a certain period of time, and the state of fatigue driving behavior will be weakened or even forgotten by the face positioning face motion curve to determine whether the driver is fatigue. (3) studied the facial feature points the traditional location algorithm, based on SDM facial feature points detection on tracking feature points of eyes, the topological structure of the human eye feature point tracking based on can We quickly get the location of human eye feature points. On this basis, a fatigue driving detection method based on PERCLOS criterion is proposed, which is used to analyze the ratio of eye width to height and the black pixel area of human eye. Experimental results show that the algorithm is of good accuracy.

【学位授予单位】：中北大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】