【作者】 刘志峰；

【导师】 李克强；

【作者基本信息】 清华大学 ， 机械工程， 2011， 博士

【摘要】 车辆横向安全预警系统能够有效避免或降低行驶碰撞带来的人员伤亡和经济损失，对提高交通安全具有重大意义。针对传统预警系统存在的横向安全预警功能集成性不足和误警率漏警率较高的问题，本文提出一种车辆集成式横向安全预警系统结构，并开展车辆驾驶意图识别、符合驾驶员习惯的车道偏离预警算法和邻车道后侧向车辆识别等关键技术的研究，实现车道偏离预警和换道碰撞预警功能的集成，并对各项关键技术进行实验验证，以达到提高系统精度及系统对多种路况和不同驾驶员的适应性的目的。论文首先研究车辆驾驶意图识别方法。通过对各种信息在大量不同驾驶员的车道保持和车道更换行为下的统计分析，确定了它们对于车辆驾驶意图的反映程度，从而提取出应用于意图识别的特征信息组，将其输入到支持向量机中，通过对支持向量机方法中内核、时窗及特征向量的研究选取，提出基于信息统计的车辆驾驶意图识别方法，有效提高了识别的准确率和实时性。然后，研究适应驾驶员习惯的车道偏离预警方法。通过对传统预警方法存在误警和漏警原因的分析，提出一种驾驶员横向驾驶行为特性指标，并进一步得到了该特性的区域划分方法，在此基础上设计考虑驾驶员横向驾驶行为特性的分区域分指标的车道偏离预警算法，并提出一个综合误警率和漏警率的指标来分析预警效果和确定算法参数，有效提高了预警的准确性。其次，研究基于邻车道后侧向车辆识别的换道碰撞预警方法。通过识别邻车道来规划车辆识别区域，选取了底部阴影、边缘和形状特征进行后向车辆识别，并提出一种基于多梯形区域的三阶累计量统计方法，在后向车辆识别方法识别失效时，实现对侧向车辆的稳定识别和跟踪，并设计预警策略，完成换道碰撞预警功能。最后，为验证所提出方法的有效性和正确性，搭建相关实验平台，并进行了大量驾驶员的实车实验。使用采集到的实验数据完成了车辆集成式横向安全预警系统关键技术的验证，结果表明，本文设计的车辆集成式横向安全预警系统能够有效区分车辆驾驶意图，采用的预警算法符合驾驶员习惯，能降低误警率和漏警率，明显改善预警效果。更多还原

【Abstract】 The vehicle lateral warning system can effectively decrease the crash accidents,make drivers easier and thus has a great significance for the traffic safety. Conventionalsystems should make some improvements in the performances due to the lack ofintegrated lateral warning functions and the high false and missed alarm rates. To solvethese problems, a scheme of vehicle integrated lateral warning system is proposed andthe key technologies are studied, which include the driver lateral operation intentionidentification method, the lane departure warning algorithm considering lateral drivingbehavior characteristics and the lane change collision warning algorithm based onrear/side vehicle detection. Thus the integration of lane departure and lane changecollision warning functions is realized and the key technologies of the integrated systemare verified to adapt to the variety of road conditions and driving styles.Firstly the driver lateral operation intention identification method is studied. Alarge amount of driver experiment data is collected and analyzed to determine therelationship with driver intentions and thus the feature data set for recognition isselected. The Support Vector Machine method is applied and the accuracy and real timeperformances of driver intention identification are improved.Then the lane departure warning algorithm considering driver lateral drivingbehavior characteristics is investigated. The cause for missed and false alarms isanalyzed and an index for lateral driving behavior characteristics is proposed. Based onthis cause, the lateral driving areas are divided and the lane departure warning algorithmdepending on area and index is determined. A warning efficiency index integrating falsealarm rate and missed alarm rate is designed to validate the warning efficiency.Meanwhile the lane change collision warning algorithm based on rear/side vehicledetection is determined. The vehicle detection area is programmed through side lanerecognition and the rear vehicle recognition is done based on multi-feature fusion. Athree order cumulant statistics method is designed based on the multi-trapezoid areas,which can realize the robust side vehicle recognition and tracking where the rear vehiclerecognition method makes no effect. The disturbances are excluded and the real-timeperformance and accuracy are improved. Finally field tests are carried out to validate the proposed methods. The test resultsshow that the proposed vehicle integrated lateral warning system can effectivelydistinguish the drivers lateral operation intentions and the warning algorithms are inaccordance with the driver characteristics, thus reducing false alarm rate and missedalarm rate and improving warning effects.更多还原