基于马尔科夫毯和隐朴素贝叶斯的驾驶行为险态辨识  

Driving risk status identification based on Markov blanket hidden Naive Bayes

作　　者：严利鑫 黄珍[4] 朱敦尧[1,2] 陈志军[1,2] 冉斌[3] 

机构地区：[1]武汉理工大学智能交通系统研究中心,武汉430063 [2]国家水运安全工程技术研究中心,武汉430063 [3]威斯康星大学麦迪逊分校土木与环境工程学院,麦迪逊53705 [4]武汉理工大学自动化学院,武汉430070

出　　处：《吉林大学学报（工学版）》2016年第6期1851-1857,共7页Journal of Jilin University:Engineering and Technology Edition

基　　金：国家自然科学基金项目(51178364;61104158);武汉理工大学教学研究项目(2011180)

摘　　要：为了实现对驾驶行为险态的有效辨识,以实时采集的多源信息为依据,通过融合驾驶人心率变化率及违法行为将驾驶行为险态分为4级。采用马尔科夫毯特征抽取算法提取出速度、纵向加速度、前轮转角变化率、车道偏离量以及车辆位置作为构建驾驶行为险态辨识的特征集,基于隐朴素贝叶斯(HNB)构建驾驶行为险态辨识模型。十折交叉验证结果表明,该模型的辨识精度(90.6%)比朴素贝叶斯(NB)、贝叶斯网络(BN)及径向基函数(RBF)神经网络分别提高14.1%、13.9%和13%。此外,ROC曲线验证结果表明该模型对不同险态等级都具有良好的预测效果。In order to effectively identify risk status while driving, a driving risk status identification model is proposed based on the information of driver operation and vehicle status. According to the rate of electrocardiogram （ECG） and traffic violation behaviors, the driving risk status is classified into four levels. Using Markov blanket algorithm, five factors are selected as the feature set, including the speed, the longitudinal acceleration, the rate of front wheel angle, the vehicle position and lane departure. Then, the algorithm of Hidden Naive Bayes （HNB） is employed to establish the driving risk status identification model. The results show that the accuracy of HNB is 90. 6%：, increasing 14.1%, 13.9 and 13.0% compared with Bayesian Network （BN）, Naive Bayes （NB）andRadial Basis Function Neural Network, respectively. The results of ROC curve indicates that the model presents high predictive power. The conclusions can provide theoretical support for designing dangerous driving status recognition equipment based on vehicle and driver operation information.

关 键 词：道路工程 险态辨识 马尔科夫毯 隐朴素贝叶斯 交通安全 

分 类 号：U491[交通运输工程—交通运输规划与管理]