【作者】 汪澎；

【导师】 刘志强；

【作者基本信息】 江苏大学 ， 车辆工程， 2010， 博士

【摘要】 道路交通事故是多年来困扰世界各国的社会问题,它直接关系到人民生命和财产的安全。随着经济的发展,车辆拥有量增加和非职业驾驶人人数增多,致使交通事故量大幅增加,已经成为当今社会第一公害,给人类社会造成的伤害大于任何一种自然与社会灾害。影响交通安全的因素有很多,其中驾驶人的人为影响因素是造成交通事故的主要原因。大量的统计分析结果表明,70%以上道路交通事故的发生都与驾驶人自身及其驾驶行为特性直接相关。其中,驾驶人低警觉性驾驶行为是影响交通安全的驾驶人为因素中最重要的组成部分,是引发恶性交通事故的重要原因之一。低警觉性驾驶行为的深入研究有助于提高驾驶人安全警惕能力,实现驾驶人安全驾驶的目的。目前国内外驾驶人行为特性的相关研究主要集中在驾驶人心理认知性、生理应激反应特性及驾驶适应性等方面。研究大多采用单因子分析法,以问卷数据调查或者实验室仿真模拟的方法进行分析,结合统计学理论分析驾驶行为的变化规律,从驾驶人心理与生理特性研究入手,挖掘引发危险驾驶行为的本质原因,为交通安全管理与行车安全保障研究提供科学依据。在特定实验室环境条件下进行的驾驶行为仿真模拟分析与实际驾驶时驾驶人身临其境条件下的心理特性、生理特性和行为特征有本质区别,在驾驶人警惕性和兴奋程度上有巨大的差异。基于既往驾驶人驾驶行为特征数据统计的研究虽然构建出很多驾驶行为数学表述模型,对驾驶行为的规范化和安全化起到了示范性指导作用,但是实际驾驶过程中驾驶人面临迥异的环境,心理和生理上的差异和驾驶行为具有突变性的特点,使得现有驾驶模型研究成果难以应用到驾驶个体安全保障的实际中去。目前,驾驶行为实时监测、分析和评价的工程实用性研究成果尚未见发表。警觉度是指人集中精力执行一项操作任务时所表现出的灵敏程度。许多人机交互系统需要操作人员保持一定的警觉度。驾驶警觉性是指驾驶人对车辆行驶刺激信息保持警惕并且能够作出适当应激反应的能力,这种能力必须贯穿于驾驶过程的始终。任何一个小小的疏忽都有可能导致交通事故的发生,这种疏忽性行为是本文研究对象——驾驶人低警觉性驾驶行为。本课题源于国家科技支撑计划课题“驾驶人安全驾驶行为分析平台与监测技术研究及示范”中的子项专题“驾驶人不安全驾驶状态监测与预警分析技术研究”,主要开展驾驶人低警觉性行为特征检测以及智能评价技术的研究。驾驶人在正常驾驶状态和警觉性降低状态时会表现出不同的行为特征,其中表征驾驶人驾驶性能的行为指标主要包括：车道安全保持的稳定性,相对车间距控制的安全性,驾驶人注视行为的安全性,车辆速度、方向控制能力,对外部事件的反应能力以及驾驶强度的高低等。目前相关研究大多基于驾驶人心电波、脑电波信息、脉搏血压和体内新陈代谢产物的监测进行车辆控速、对外部事件的反应能力以及驾驶强度的研究,该方法虽然准确直接,易于做出正确的评价和判断,但是接触式的测量且仪器复杂,难以进行车载实时在线检测应用。针对目前存在的问题,研究以驾驶人低警觉性不安全驾驶行为为研究对象,构建驾驶人驾驶行为实时监测分析平台,针对驾驶个体进行驾驶行为组织结构化和过程化的研究,由个体到群体,由表及里进行驾驶低警觉性行为因子的检测和融合评价。研究采用实车试验的方法,设定多种不安全驾驶工况,以驾驶人状态和车辆行驶状态表征为检测内容,实时对驾驶人个体进行测试,实现驾驶人低警觉性行为表征信息的提取,完善了驾驶行为检测分析技术。研究深化了对驾驶人因素引发事故机理的研究,把握不安全驾驶行为的规律、特点和深层次原因,拓宽驾驶人不安全驾驶行为机理研究的深度与广度。驾驶低警觉性驾驶行为表征信息具有多源性,除通过医学检验反映出的驾驶人反应能力和驾驶强度指标外,还包括不安全驾驶注视表征,车道偏离表征(含方向控制能力),以及危险车间距表征(含速度控制能力)等信息。研究应用机器视觉检测技术对驾驶人进行实时监测,信息主要包括两个部分：驾驶人注视特征信息和车辆行驶外在表征特征信息。通过实时获取车道偏离信息和相对车间距信息,实现车道偏离状态和危险车间距跟驰的检测。研究通过实时检测的驾驶人脸部特征信息进行驾驶人注视特征信息提取,实时提取驾驶人双眼间的横向宽度(瞳距)和嘴巴到双眼连线中点之间的纵向距离——人脸“T”型线信息,实现驾驶人不安全注视特征状态的检测。研究运用多智能体理论构建低警觉性驾驶行为监测体系,运用数据挖掘技术从大量检测数据中提取出低警觉性驾驶行为的特征信息。采用贝叶斯与神经网络技术进行特征模式分类,运用Dempster-Shafer证据推理技术进行信息融合,实现多智能体系统协商合作和策略规则匹配,实现低警觉性驾驶行为的智能决策,提高低警觉性驾驶行为表征信息甄别的自适应性、准确性和智能化水平。更多还原

【Abstract】 Traffic accident, which directly relates to the loss of both people’s lives and prosperities, is one of the problems concerned by the whole world. In recent years, the traffic accidents arise more and more frequently along with the growth of the number of cars and drivers, causing a lot more damages to people, society and world than any other form of natural and social disaster. Quantities of statistics show that more than 70 percent of road accidents are directly related with drivers and their driving behavior characteristics. Every year on the high-way, driver’s low-alertness behavior is one of the important causes of serious accidents. The research is about to improve the driving features, ensuring the driving security and prevent the traffic accidents.The researches of driving behavior characteristics mainly focus on the driver’s psychology cognitive function, physical stress response characteristics, driving adaptability and so on. Driver’s physiology and psychology are mostly studied through the single factor analysis with the questionnaire data investigation and the laboratory simulation analysis. The changing rule of the driving behaviors is studied through statistical analysis and essential reason causing dangerous driving is studied through the driving behaviors’characteristics, providing scientific basis for traffic safety management.In particular the laboratory environment conditions of the simulation research and practical driving while driving the scene has essential difference in driving vigilance and motorists excited extent have bigger difference. The mathematical representative model, which is based on previous drivers driving behavior characteristic data statistics, certainly could conduct the driving rule. But when it comes to the application to the security of a driver personal, the model can hardly satisfy the practical application, giving consideration to the difference of the driver’s physiology, psychology and the environmental varieties. The positively engineering practical research results about the monitoring and analyzing on driving behaviors in real time haven’t been reported yet.Alertness degree is used for depicting the sensitivity when people focus on operating. People-machine interactive system needs people to keep focusing. For example, a driver needs to keep sensitivity when he drives in the highway. Driving sensitivity is a capability of responding effectively to the driving information without discontinuity. One tiny mistake may lead to a traffic accident. This inattentive behavior-driving in low alertness, is the research center of this paper. This subject is from the subject-research on the monitoring and warning analysis technology about drivers’unsafe driving behaviors, which is a sub-subject of the national science and technology support program subject-safe driving behavior analysis platform and monitoring technology and demonstration project. According to the existing problem, this study focuses on the research of the accidental mechanism causing by the drivers, the unsafe driving behavior rule, the characteristics and the deep reasons, broadening the depth and breadth of the mechanism of the safe driving behavior.Drivers in normal driving and alertness lower have different driving behavior characteristic: lane departure, dangerous relative distance between cars, the driver watch characteristics, the speed and direction controlling, the reaction of the driver and the working intensity. Based on the heart waves information, brain wave information, pulse pressure, blood pressure and metabolic product monitoring, these methods is accurate and directly to the correct evaluation and judgment. But contact measurement and complex instrumentations are hard to use in the vehicle real-time online. Aiming at the existing problems, the low-alertness driving behaviors are the research subject in this paper. Construct a real-time monitoring and analyzing platform for safe driving behaviors through influencing factors of the driving behaviors. Make the detection analysis technique on driving behaviors better from individual to the group, from point to surface and from outside to inside, according to the organizational structure and process of the driving behaviors. The experiment is set for different unsafe conditions, detecting the state of the driver and the vehicle including different drivers.The driver’s low-alertness behavior characteristics is very complicated, including information of the lane departure, the position and velocity of other vehicles and other targets, the state of the driver and so on. The information of the driver, which is got by machine vision, consists of two parts:the information of the driver’s awareness and the vehicular exterior characteristic information. The lane departure state and the dangerous vehicle-following distance are detected through real-time information about the departure and the relative distance between vehicles. Researches about real-time detections of facial features of drivers are extracted through the horizontal width of the eyes and the longitudinal distance between the mouth (pupil) to the eyes in the midpoint-the "T" type face information safety features. These are for the detections of the distraction state.Construct the low-alertness driving monitoring system based on multi-agents theory and extract the feature information about the low-alertness driving behaviors from quantities of statistics. The feature is classified by Bayesian and neural technique. The match of the multi-agents system and the strategic rule is based on Dempster-Shafer evidence theory, estimating the low-alertness driving behaviors intelligently and improving the adaptability, accuracy and intelligent of the representation on the low-alertness driving behaviors.更多还原