【作者】 黄勇；

【导师】 孙棣华；

【作者基本信息】 重庆大学 ， 控制科学与工程， 2019， 硕士

【摘要】 驾驶疲劳识别是改善道路交通安全的重要途径之一。现有基于驾驶员外部特征的识别方法容易受到光照等环境因素的影响,而基于生理特征的识别方法受限于高昂的设备成本以及对驾驶员的入侵性等主要还是应用于医学领域。近年来,基于驾驶员操作和车辆状态的识别方法逐渐受到众多研究者的重视,能够有效避免前两种方式的不足,但是该方法需要在车辆上安装多种传感器,提取的部分疲劳特征在实际中不易获得,大多还处于模拟驾驶研究阶段。为此,研究一种不依赖车载传感器的驾驶疲劳识别方法,对于实现疲劳驾驶监测与控制具有重要实际意义。考虑到当前智能手表和手机已经逐渐普及,它们所带的传感器能够有效监测物体的运动,本文基于驾驶员操作和车辆状态的识别方法,采用智能手表和手机分别监测驾驶员的手部操作行为以及车辆运动状态,并围绕不同驾驶状态下两种智能设备所采集数据的特征分析、有效表征不同驾驶状态的疲劳特征提取、最优疲劳特征组合筛选、疲劳识别模型的建立及改进等方面展开研究,建立了一种更加便捷实用的驾驶疲劳识别方法。主要内容包括:(1)考虑到不同的数据信息存在不同的疲劳特征,本文通过对智能手表和手机传感器数据的波动差异性进行分析后,采用时域特征法、频域特征法和样本熵特征法三种方式从这些传感器数据中提取了多种特征指标,并采用单因子方差分析对这些指标进行显著性检验,从而量化特征指标的有效性并得到多个有效疲劳特征指标。(2)针对实车情形下车辆行驶状态信息对智能手表所采集的驾驶员转向操作信息的影响,本文利用智能手机所采集到的车辆行驶状态信息,基于三维坐标系旋转方法设计了实车情形下从驾驶员转向操作信息中分离掉这部分车辆行驶状态信息的方法,为实车情形下能够提取出有效疲劳特征奠定了基础。(3)为了建立具有高准确率的疲劳识别模型,分别以支持向量机、RBF神经网络和K近邻模型的分类准确率建立指标优选的评价准则函数,将智能手表和手机传感器数据中所提取疲劳特征指标结合并使用序列前向浮动选择算法对其进行优选搜索,从中选择识别效果最好的优选特征集和分类模型建立疲劳识别模型。在此基础上,进一步分析了驾驶员差异对模型的影响,从而对模型进行了改进。更多还原

【Abstract】 Driving fatigue recognition is one of the important ways to improve road traffic safety.In the existing methods,the driver external characteristics-based method is easily affected by environmental factors such as illumination,and the physiological characteristics-based method is mainly applied in medical field because of its high equipment cost and intrusiveness to drivers.In recent years,the driver operation and vehicle state-based method has been paid more attention by many researchers,which can effectively avoid these shortcomings of the former two.However,this method needs to install a variety of sensors on the vehicle.Some of the fatigue features are not easy to obtain in practice,and most of researches are still in the stage of simulation research.Therefore,it is of great significance to study a driving fatigue recognition method for realizing fatigue driving monitoring and control that does not rely on on-board sensors.Considering that smartwatch and smartphone are now widespread and can effectively monitor the movement of objects,based on the driver operation and vehicle state-based method,this paper uses smartwatch and smartphone to collect drivers’ hand behavior and vehicle state data respectively.A more convenient and practical method of driving fatigue recognition is established by analyzing data feature of two devices under different driver states,extracting effective fatigue features representing different driver states,selecting the optimal combination of fatigue features,establishing and improving the fatigue recognition model.The main research work of this paper includes the following three aspects:Firstly,considering that different data information has different fatigue features,after analyzing the fluctuation difference of smartwatch and smartphone sensors data,this paper extracts a variety of feature indexes from these sensors data by three ways: time domain feature method,frequency domain feature method and sample entropy feature method,and uses one-way ANOVA to test the significance of these indexes.Several effective fatigue feature indexes are obtained after the validity of the feature index is quantified.Secondly,aiming at the influence of vehicle driving state information on drivers’ steering operation information collected by smartwatch in real-vehicle,based on threedimensional coordinate system rotation method,this paper designs a method of separating this part of vehicle driving state information from driver steering operation information in real-vehicle by using the vehicle driving state information collected by smartphone,which lays a foundation for extracting effective fatigue features in real-vehicle condition.Thirdly,in order to establish a fatigue recognition model with high accuracy,all of the fatigue features extracted from smartwatch and smartphone sensors data are used to optimally select by sequential forward floating selection algorithm,and the evaluation criterion function of feature optimization is established by the classification accuracy of support vector machine,RBF neural network and K-nearest neighbor models respectively.The fatigue recognition model is established based on the classification model and optimal features with the highest accuracy.On this basis,the influence of driver differences on the model is further analyzed,and the model is improved.更多还原