面向驾驶员疲劳车道偏离识别方法研究

发布时间：2018-01-04 05:33

本文关键词：面向驾驶员疲劳车道偏离识别方法研究　出处：《吉林大学》2017年博士论文　论文类型：学位论文

【摘要】：对近几年道路交通事故统计资料的分析发现,疲劳驾驶是导致重特大交通事故的主要因素之一.因此,运用各种检测手段对驾驶员的疲劳状态进行检测已成为近几年的研究热点.在交通事故发生前进行报警并采取相应措施,对于降低交通事故的发生率,减少人员伤亡及财产损失具有重要的社会意义和经济意义.本文提出了考虑驾驶员操纵特性与车辆运动状态特性的疲劳车道偏离识别方法,并做出驾驶员是否疲劳的判断,在满足科研团队国际合作项目需求的基上,为车道偏离预警系统决策提供一定的理论支持.主要工作如下:本文基于吉林大学汽车仿真与控制国家重点实验室的驾驶员在环试验台设计了疲劳车道偏离试验.试验以12名驾驶员为研究对象,采集了能够反映驾驶员疲劳车道偏离的操纵特性与车辆运动状态特性的相关参数.然后借助受试者工作特性曲线(ROC)确定能够明显区别驾驶员正常换道和疲劳车道偏离的识别时间窗.最终确定驾驶员车道偏离识别时间窗为3s.以此筛选有效样本,并将样本分为训练样本和识别样本.运用独立样本T检验,分别量化了疲劳偏离和正常换道状态对驾驶行为特性参数的影响,并提出了能够分别疲劳车道偏离和正常换道的特性参数.以驾驶行为特征参数作为观测层,建立高斯-隐马尔科夫疲劳车道偏离识别模型.为了对比不同特征参数类型对识别效果的影响,模型建立过程中采用了10类特征参数集分别对模型进行了离线训练,得到了相应的模型参数.为了对比不同建模方法对识别效果的影响,我们进行下面两个步骤:步一:我们建立了两类驾驶疲劳识别模型是基于支持向量机(SVM)的疲劳车道偏离识别模型和基于方向盘转角速度时序分析的驾驶疲劳识别模型.首先,针对上面10类特征参数集,基于SVM理论建立了疲劳车道偏离识别模型,用于分别疲劳车道偏离及正常换道的状态.其次,利用上面已确定的识别时间窗,选取时间窗内的方向盘转角速度数据序列作为识别特征.当识别特征满足波动幅度约束与波动变化约束时,则认定该操作时段驾驶员存在疲劳.步二:通过模型识别效果分析,研究了特征参数类型及不同建模方法对识别效果的影响.结果表明:从准确率、灵敏度和特异性三个模型评价函数综合分析得出,与所建立的模型相比,基于GM-HMM建立的疲劳车道偏离识别模型识别效果最优.说明,研究成果能够为汽车主动安全辅助系统的研究和应用提供一定的理论和技术支持.
[Abstract]:The analysis of statistical data in recent years, the road traffic accident, driver fatigue is one of the main causes of serious traffic accidents. Therefore, using various means of detection of driver fatigue detection has become a hot research topic in recent years. When the traffic accident happened before the alarm and take corresponding measures to reduce traffic accidents. The incidence rate has important social and economic significance to reduce casualties and property losses. In this paper, considering the fatigue characteristics and driving lane vehicle motion deviation recognition method, and make the judgment whether the driver fatigue, to meet the needs of the project research team of international cooperation based on departure warning system to provide a decision the theoretical support for the lane. The main work is as follows: the Jilin University State Key Laboratory of automotive simulation and control based on driving The driver in the loop test platform is designed to test fatigue test. Lane departure 12 drivers as the research object, collected can reflect the related parameters of driver fatigue characteristics and operating characteristics of lane vehicle motion deviation. Then by means of the receiver operating characteristic curve (ROC) is able to identify significant differences for the driver's normal time window road and lane departure. Ultimately determine the fatigue of driver identification lane departure time window for 3s. to screen the effective sample, and the sample is divided into training samples and identifying samples. Using independent sample T test, respectively to quantify the fatigue and deviate from the normal state change impact on driving behavior characteristic parameters, and puts forward the characteristic parameters to lane departure and normal fatigue respectively. The lane changing driving behavior characteristic parameters as the observation layer, a Gauss Markov model. In order to identify the fatigue of lane departure Effects of different types of feature parameters than the recognition results, the process of modeling, using 10 kinds of characteristic parameter sets were carried out off-line training of the model, the corresponding parameters are obtained. In order to compare different modeling methods on the identification results, we carried out the following two steps: step one: we set up two kinds of driving fatigue recognition model is based on support vector machine (SVM) fatigue recognition model of lane departure and driving fatigue recognition model analysis of steering wheel angle velocity based on time series. Firstly, according to the above 10 kinds of feature parameters set, SVM theory established the fatigue recognition model based on lane departure, respectively for lane departure and normal fatigue state change. Secondly, using the recognition time window above have been identified, selected within the time window of the steering wheel angular velocity data sequence as the recognition feature. When the identification feature satisfies wave amplitude constraint Constraint and fluctuation, is that the operation time. Step two: driver fatigue identification effect through the model analysis, studied the effect of parameters and different types of modeling methods on the identification results. The results show that the accuracy, sensitivity and specificity of the three models on comprehensive analysis of price function that, compared with the established the optimal GM-HMM model, established from fatigue lane recognition model based on recognition results. That research results can provide some theoretical and technical support for the research and application of vehicle active safety auxiliary system.

【学位授予单位】：吉林大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：U463.6

【相似文献】