基于生理信号的驾驶员疲劳检测方法研究

Fatigue driving is a significant cause to road traffic accidents, and a driver’s fatigue state could be reflected in physiological signals in which a large amount of information is contained. In the current study, a driver’s fatigue detection model is usually established based on signal features picked up by time domain or frequency domain methods from a single type of physiological signal. Besides, most of the existing research is completed in a low-level driving simulator with a small sample size, and this situation lead to a poorly reproduced working condition as well as experimental results with low reliability. The proposal is aimed to study the fatigue state of drivers by acquiring all kinds of physiological signals, and extracting signal features using time domain, frequency domain, time-frequency domain and nonlinear methods as many as possible, combined with the forefront pattern recognition methods such as deep learning and Rodriguez’s method. At the same time, the proposal is planned to be completed in a high-level driving simulator with a large sample size. In this way, the problem of the singleness of physiological signal type employed, the limitation of signal feature extraction methods used, and the obsolescence of pattern recognition methods involved, which are exit in the relative researches in recently years, could be overcome. And therefore, the reliability and repeatability of the research results could be ensured. The proposal intends to establish an accurate and non-invasive driver’s fatigue detection model, as well as provide a theoretical basis and the data support for the study of the mechanism of fatigue by studying the relationships between different features from various physiological signals and fatigue, and the relationships between different features themselves.

疲劳驾驶是造成道路交通事故的重要原因之一，而生理信号中蕴藏的信息能够反映驾驶员的疲劳状态。目前的研究多选用一种生理信号，采用时域或频域方法提取其中的特征，建立疲劳检测模型，模型性能仍有许多不足之处。另外，现有研究通常采用小样本量在简单型驾驶模拟器上完成，实验过程对实际工况的再现能力不足，实验结果可靠性不高。本项目提出同步采集各种生理信号，全面提取信号中的时域、频域、时频和非线性特征，并结合深度学习和Rodriguez法等最新模式识别方法，研究驾驶员的疲劳状态，同时采用大样本量在高性能驾驶模拟器上完成，由此克服近年来相关研究存在的选用生理信号类型单一、信号特征提取方法不多，模式识别方法过时等问题，确保研究结果的可靠性和可重复性。本项目拟在研究各种生理信号不同特征与疲劳的关系，以及不同特征之间相互关系的基础上，建立准确、无创的驾驶员疲劳检测模型，并为疲劳机制的研究提供理论依据和数据支持。

驾驶疲劳的研究是交通安全领域的一个重要课题。本项目通过3批次驾驶疲劳实验，研究了驾驶员疲劳状态的生理特征和操作特征。首先，分析了长时间单调驾驶作业开始时和结束时的心率变异性信号和脑电信号。结果表明，与额叶位置相比，颞叶位置脑电变化更为显著，对用于困意驾驶的可穿戴检测设备而言，颞叶位置是可行的脑电记录位置；通过去趋势波动分析提取心率变异性信号和脑电信号中的标度指数，结果表明这一特征具有应用于驾驶疲劳检测系统的前景。其次，研究了驾驶员心率变异性特征与精神状态的关系。结果表明心率变异性特征可以表征驾驶员的精神状态，并且在表征驾驶员精神状态时存在性别差异；男性和女性驾驶员自主神经系统的活动在疲劳状态存在更多的性别差异。最后，提取了驾驶员疲劳驾驶时的操作特征，并采用联合优化方法对分类器参数与指标集进行优化，通过支持向量机建立疲劳检测模型，准确率达到81.82%。本项目的研究成果为驾驶疲劳检测系统的开发提供了理论依据和数据支持。

内容获取失败，请点击重试