【作者】 王亚丽；

【导师】 金立生；

【作者基本信息】 吉林大学 ， 载运工具运用工程， 2013， 硕士

【摘要】 随着汽车保有量的增加，道路交通事故的频繁发生已成为全球性的重大经济和社会问题。其中车和车追尾碰撞是高等级公路最易发生的交通事故。因此为了有效减少或避免此类交通事故的发生，需研发一种前方车辆检测系统，当车辆前方出现有效目标车辆时，能及时准确地给驾驶人提供警告，并能获取车辆类障碍物相对于本车的位置、速度、加速度等状态信息，从而为车辆行驶前方障碍物监测预警系统的研发奠定理论和技术基础。本文提出了基于毫米波雷达与机器视觉信息相融合的前方车辆检测方法，利用毫米波雷达获取前方有效目标车辆的距离、速度和加速度等信息；依据车辆几何特征信息及毫米波雷达提供的距离信息在摄像机图像上建立感兴趣区域；然后对感兴趣区域利用Adaboost算法进行验证，判断是否为车辆类障碍物。主要研究内容如下：1、基于毫米波雷达的有效目标确定。首先，将毫米波雷达获取的障碍物信息按距离原则进行有序排列；在分析测量环境的基础上，通过设定距离阈值提取出同车道目标，使用距离最近原则初选出有效目标；将目标预测信息与初选目标信息进行一致性检验；使用有效目标生命周期的方法进行目标决策，实现有效目标的准确选择。2、基于机器视觉的车辆识别方法。在采集大量正负样本的基础上，对样本进行灰度化、归一化处理，建立20×20像素的图片库，从而构成训练样本；利用积分图快速的提取出Haar-like矩形特征集；基于Adaboost从Haar-like特征集中提取出有效特征生成同等数量的弱分类器，最终将弱分类器加权组合成强分类器，其中利用权值体现弱分类器的性能。最后采用Adaboost与级联分类器相结合的方法将强分类器构成前方车辆检测分类器。3、基于毫米波雷达与机器视觉融合模型的搭建。首先进行两传感器的数据融合，主要包括：空间融合和时间融合。反用基于单帧静态图像的测距模型，建立了坐标转换关系，实现数据在空间上的融合。基于张正友标定原理求出了摄像机的畸变参数，对空间上坐标转换关系进行了修正。利用多线程方法实现了数据在时间上的融合。在搭建数据融合模型的基础上，利用毫米波雷达获取的距离信息，将前方目标车辆投影到摄像机图像上建立感兴趣区域；基于搭建的前方车辆检测分类器，对目标车辆有效性进行验证。运用VC6.0开发基于毫米波雷达与机器视觉信息相融合的车辆检测系统。对前方车辆检测分类器在PETS（Performance Evaluation of Tracking and Surveillance)提供的图片上进行测试，并将融合系统在真实道路环境条件下进行实车实验。实验结果表明，该方法可以快速、准确地实现前方车辆的在线检测，具有较好的实时性和准确性。更多还原

【Abstract】 With the increasing quantity of automobile, the road traffic accident occurs so frequently that ithas become one of the major global economic and social issues. Rear-end collision in highway ismakes up the most part of traffic accidents. To avoid or reduce such traffic accidents, a system ofvehicle detection should be developed. When other vehicles appear in the front of the vehicle asobstacles, the system can warn the driver accurately in real time. And the system can get theinformation including distance speed and acceleration between the vehicle and the obstructivevehicle, which provide information to support the vehicle assistance system.Vehicles were detected by means of vision and millimeter-wave radar data fusion. Themillimeter-wave radar provides a first estimation of vehicle candidates with related lateral positionand distance information. The information is used to define a region of interest (ROI) on an imageobtained by the vision. An Adaboost object detection algorithm is utilized to scan the ROI andverify the detection of millimeter-wave radar.1.Determine the dangerous target based on the millimeter-wave radar. Firstly, objects wereranged according to the distance principle. Then the nearest object in the same lane was determinedas the first selection based on analysis of the measurement environment, and the target determinationwas then completed through a relevant target life cycle method.2.Vehicle detection based on the vision. It was trained using vehicle rear views andnegative-samples as training images. The image library was generated by graying method andnormalized method. Haar-like features were proposed by integral images. Combining Adaboostalgorithm the most important ones were selected.3.Data fusion method based on Millimeter-Wave Radar-vision. Data fusion method include twoaspects: data fusion method on space and in time. Build the data transmission of coordinate withdistance measurement model based on single motionless image. The distortion parameters wereacheived in the way of Zhengyou Zhang’s calibration principal. Data fusion in time was finished bythe multithread system.The software system and part of the algorithm which is used for vehicle detection and recognition have been developed based on Visual6.0. Pictures of IEEE International Workshop onPerformance Evaluation of Tracking and Surveillance (PETS) were provided to test the detectionsystem of ahead vhicle. The results of real image experiments have proved to be satisfactory.更多还原