基于心电信号的驾驶疲劳识别方法研究

发布时间：2018-10-25 16:34

【摘要】：随着车辆保有量的上升,驾驶安全日益引起人们的重视。驾驶员作为交通系统的重要组成部分,其驾驶状态直接决定了整个交通系统的安全水平。而在众多影响驾驶员状态的因素中,驾驶疲劳最为常见,因此,如何对驾驶员的驾驶状态进行实时有效的检测,并在疲劳发生时及时发出预警,对于提升交通安全水平,降低事故率具有重大意义。针对上述问题,本文在充分借鉴现有研究成果的基础上,构建了一种基于心电信号的驾驶疲劳识别模型。本文的主要研究内容包括以下几点:1.首先介绍了驾驶疲劳检测的背景及研究意义,分析国内外相关研究在研究对象、研究方法和研究结论上的最新进展,随后提出本文的研究内容和技术路线。2.从认知心理学角度阐述了驾驶疲劳产生机理,分析了驾驶疲劳的诱发因素及疲劳表征。详细介绍了心电信号的处理方法和指标提取理论,通过对心率变异性和R-R间期的分析,确定了 R-R间期的提取方法,并结合数据对该方法进行了验证,为进一步提取可有效表征驾驶员生理状态的心电指标奠定了理论基础。3.设计了双任务范式的长时间模拟驾驶实验,主任务为车辆跟驰,次任务是对刹车信号的按键反应。通过分析实时采集的被试心电和行为数据后发现,随着疲劳的发生,被试心电指标和行为指标均呈现一定的趋势性,且经过显著性分析,实验前期和后期,大部分心电指标差异显著,因此,可初步判断出对驾驶疲劳指向性较好的心电指标。4.阐述了将反应时间作为疲劳等级划分依据的可行性。通过分析整个实验过程中简单反应时间的变化规律,提出了利用反应时间划分疲劳等级。首先将实验过程划分为若干时段,以第1时段的驾驶状态为轻度疲劳状态,通过显著性分析,对剩余时段的驾驶状态进行标定,区分轻重两种疲劳状态。此外,通过与反应时间的相关性分析,提取了可有效反映疲劳状态的心电指标,构成疲劳识别心电指标集。5.利用SVM理论构建了疲劳识别模型,通过不断调整疲劳识别心电指标集构成和核函数,对模型的识别效果进行分析,发现综合选用时域和频域指标以及RBF核函数时,模型的识别效果最优。最后利用实验数据验证了上述模型的有效性。
[Abstract]:With the increase of vehicle ownership, people pay more and more attention to driving safety. As an important part of traffic system, driver's driving state directly determines the safety level of the whole traffic system. Driving fatigue is the most common among the many factors that affect driver's condition. Therefore, how to detect driver's driving state in real time and give out early warning in time when fatigue occurs can improve the level of traffic safety. It is of great significance to reduce the accident rate. In order to solve the above problems, a driving fatigue identification model based on ECG signal is constructed based on the existing research results. The main contents of this paper include the following: 1. This paper first introduces the background and significance of driving fatigue detection, analyzes the latest progress in the research object, research methods and research conclusions at home and abroad, and then puts forward the research content and technical route of this paper. 2. From the perspective of cognitive psychology, this paper expounds the mechanism of driving fatigue, and analyzes the inducing factors and fatigue characteristics of driving fatigue. The processing method and index extraction theory of ECG signal are introduced in detail. Through the analysis of heart rate variability and R-R interval, the extraction method of R-R interval is determined, and the method is verified with data. It lays a theoretical foundation for the further extraction of ECG indexes which can effectively represent the physiological state of drivers. 3. A long time simulation driving experiment with dual task paradigm is designed. The main task is to follow the vehicle and the second task is to respond to the brake signal by keystroke. After analyzing the ECG and behavior data collected in real time, it was found that with the occurrence of fatigue, the ECG and behavioral indexes of the subjects showed a certain trend, and after significant analysis, the early and late stages of the experiment. There are significant differences in the majority of ECG indexes, so we can preliminarily judge the ECG index which has good directivity to driving fatigue. 4. 4. The feasibility of using reaction time as the basis of fatigue grade classification is expounded. By analyzing the variation rule of simple reaction time in the whole experiment process, it is put forward that the fatigue grade is divided by reaction time. Firstly, the experiment process is divided into several periods, and the driving state of the first period is regarded as mild fatigue state. Through the significant analysis, the driving state of the remaining period is calibrated to distinguish the heavy and heavy fatigue states. In addition, through the correlation analysis with the reaction time, the ECG indexes which can effectively reflect the fatigue state are extracted, and the set of ECG indexes for fatigue identification is constructed. The fatigue identification model is constructed by using SVM theory. By continuously adjusting the composition of ECG index set and kernel function of fatigue identification, the recognition effect of the model is analyzed. It is found that when the time domain, frequency domain index and RBF kernel function are synthetically selected, The recognition effect of the model is optimal. Finally, the validity of the model is verified by experimental data.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U491.25

【参考文献】