基于机器视觉和握力波动的疲劳驾驶研究

发布时间：2018-11-28 14:28

【摘要】：在2010年,世界的车辆总数已经超过十亿,其中美国2.4亿辆、中国7800万辆。伴随车辆的快速增长随之而来的是交通事故的增加。而研究表明其中大约20%的事故与疲劳驾驶有关。目前关于驾驶疲劳检测的研究方法基本都是应用视觉信号和脑电、心电、眼电等生理信号。由于其非接触性,视觉检测仍然是主流。但是视觉检测目前的研究有一些不足,特别是测试者在复杂环境,例如戴眼镜或者在光线变化较大,传统的二值分割等方法准确度将大打折扣。有文献表明握力变化也是衡量驾驶疲劳的有效特征之一。因此本文提出将视觉信号和握力信号融合来检测驾驶疲劳,实验结果表明相对于单一视觉信号,融合之后的算法检测更加准确。本文主要工作如下:(1)模拟驾驶平台和疲劳驾驶检测系统的搭建。为了尽可能真实地模拟实际驾驶操作,保证疲劳检测算法有效性。在硬件方面采用的模拟方向盘和脚踏板与真实的驾驶体验几乎相同,驾驶场景模拟系统的设计涵盖各种天气和路况。在此基础上构建摄像头和握力检测于一体的疲劳驾驶检测系统。(2)经压力传感器采集到的握力原始信号表征能力非常有限,首先要对原始的数据进行转换和滤波,其中滤波采用实时性更好的线性动力系统模型平滑方法。预处理之后在握力的时域特征中选择方差作为握力特征。(3)在人脸定位中研究Cascade结构的人脸检测器算法,应用基于Haar特征的Adaboost算法使得人脸定位更加准确。所有的视觉处理算法包括Adaboost和改进的主动形状模型均在基于Visual Studio2010的OPENCV中实现。(4)在定位人脸之后,利用改进的主动形状模型来进行人脸77个特征点的搜索。利用眼睛和嘴部的特征点的高宽比来定义眼睛闭合度和嘴巴张开度。最后分别在戴眼镜和不戴眼镜的测试者验证了算法的精确性和鲁棒性。(5)提出了基于人脸和握力特征融合的疲劳驾驶检测算法。对于眼睛闭合度、嘴部张开度及握力方差三个特征输入进行模糊推理系统的设计。在满足规则的交叉性和一致性方面做了大量的实验去调整模糊规则。在MATLAB中和已有的算法进行了对比仿真,证明了算法的有效性。
[Abstract]:In 2010, the world had more than one billion vehicles, including 240 million in the United States and 78 million in China. With the rapid growth of vehicles, traffic accidents increase. Studies show that about 20% of these accidents are related to fatigue driving. At present, the research methods of driving fatigue detection are based on the use of visual signals and physiological signals such as EEG, ECG and eye electricity. Because of its non-contact, visual detection is still the mainstream. However, there are some shortcomings in the current research of visual detection, especially in complex environments, such as wearing glasses or changing light, the accuracy of traditional binary segmentation methods will be greatly reduced. Some literatures show that the change of grip strength is also one of the effective characteristics of measuring driving fatigue. Therefore, this paper proposes the fusion of visual signal and grip force signal to detect driving fatigue. The experimental results show that the fusion algorithm is more accurate than the single visual signal. The main work of this paper is as follows: (1) the building of simulated driving platform and fatigue driving detection system. In order to simulate the actual driving operation as truthfully as possible and ensure the validity of fatigue detection algorithm. The simulated steering wheel and pedals used in the hardware are almost the same as the real driving experience. The driving scene simulation system is designed to cover all kinds of weather and road conditions. On this basis, a fatigue driving detection system with camera and grip force detection is constructed. (2) the original signal of grip strength collected by pressure sensor is very limited. First, the original data must be converted and filtered. Among them, filtering adopts linear dynamic system model smoothing method which is better in real time. After preprocessing, the variance is selected as the grip force feature in the time domain feature of grip force. (3) face detector algorithm based on Cascade structure is studied in face location, and Adaboost algorithm based on Haar feature is used to make face location more accurate. All the visual processing algorithms, including Adaboost and the improved active shape model, are implemented in OPENCV based on Visual Studio2010. (4) after locating the face, the improved active shape model is used to search the 77 feature points of the face. Eye closure and mouth opening are defined by the aspect ratio of the feature points of the eye and mouth. Finally, the accuracy and robustness of the algorithm are verified by the testers with and without glasses. (5) A fatigue driving detection algorithm based on face and grip feature fusion is proposed. The fuzzy inference system is designed for eye closure, mouth opening and grip force variance. A lot of experiments have been done to adjust the fuzzy rules in the aspect of meeting the intersection and consistency of the rules. The effectiveness of the algorithm is proved by comparing and simulating the existing algorithms in MATLAB.
【学位授予单位】：东华大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U463.6;TP391.41

【相似文献】