【作者】 曾杰；

【导师】 郑玲；

【作者基本信息】 重庆大学 ， 车辆工程， 2017， 硕士

【摘要】 针对毫米波雷达在智能驾驶辅助系统应用过程中,雷达数据存在着较高的虚假预警率、目标漏检的问题,本文提出一种机器视觉与毫米波雷达数据并行信息融合的日间前方车辆检测算法。首先基于雷达扫描车道和车速阈值的范围剔除一部分无效目标,并采用本文提出一种有效车辆目标滤波算法,进一步消除雷达数据中由于自身工作原因和外部干扰信号产生的无效虚假目标,以获得雷达检测车辆的在图像上投影的ROI(region of interest)区域。其次以车辆底部阴影位置初步定位车辆可能存在的ROI区域,分析车辆存在的ROI区域特点,使用Canny算子检测算ROI区域内的轮廓特征,采用ROI区域图像轮廓对称度以判断车辆存在性。最后通过视觉辅助雷达检测车辆的信息融合方式,将雷达与基于图像特征的视觉车辆检测ROI区域数据关联,以判断雷达检测ROI区域车辆存在性,并使用Haar特征分类器对雷达与视觉检测的不关联ROI区域进行复检以判定车辆存在性。本文基于VS2010 C++平台搭建起雷达与机器视觉信息融合的车辆检测算法程序,研究如何充分挖掘毫米波雷达与机器视觉传感器单独对车辆检测能力,并进一步研究了基于图像特征和基于Haar特征分类器两种算法在车辆检测方面的性能特点,最后深入研究和分析雷达与视觉信息融合在消除雷达数据存在的车辆目标漏检和误检的具体应用问题。研究结果表明:采用毫米波雷达目标滤波算法能够剔除大量无效目标,但由于雷达内部和外部干扰,存在少量目标漏检和误检的问题。基于图像特征算法检测车辆的执行效率高于基于Haar特征分类器算法,基于Haar特征分类器与基于图像特征融合的车辆检测算法可以有效降低机器视觉的误检率和提高检测正确率。基于雷达与机器视觉数据并行信息融合算法,能够有效弥补雷达误检、漏检的缺点,比传统的基于单一传感器车辆检测算法具有更好的检测正确率和较低的漏检率,能够为车辆驾驶辅助系统提供较为准确的近距车辆检测结果。更多还原

【Abstract】 Aiming at the application of millimeter wave radar in the intelligent driving assistant system,there are some problems such as high false alarm rate,missed target.An information fusion of the front vehicle detection algorithm based on the parallel association of millimeter wave radar and machine vision on daytime is proposed in this paper.Firstly,both radar scanning speed and lane range threshold are used to eliminate most of the invalid targets,and an effective vehicle target filtering algorithm is proposed for further eliminating invalid radar data,which is caused by false target jamming signal generation and external reasons.The ROI of radar data is obtained by the projecting of radar data on vision.Secondly,according to the characteristics of vehicle shadow feature in any environment,the initial ROI position of the vehicle is localized by the shadow at the bottom of the vehicle,and then contour features in the ROI region are obtained by Canny edge detection algorithm,and an image contour symmetry algorithm is introduced to estimate the existence of vehicles in the ROI region.Finally,the existence of vehicle based on the data fusion of radar and vision detection is realized by data association algorithm of two ROI regions,which are provided by image feature detection and radar detection separately.Moreover,Haar feature classifier will be used to determine the vehicle existence of none associated ROI region.In this paper,the vehicle detection algorithm based on the VS2010 C++ platform is established.Researching on how to fully develop vehicle detection ability of millimeter wave radar and machine vision,and further studying the performance characteristics of vehicle detected by image features and feature classifier algorithms.Finally,researching and analysis of radar and visual information fusion in the specific application problem of eliminating vehicle targets lost and false detection of radar.The results show that the filter algorithm for radar data can eliminate a large number of invalid targets,but the radar data still contains some lost detection and false detection targets,which are caused by the internal and external interference.Vehicle detection algorithm based on Image features is more efficient than the Haar feature classifier algorithm,and the combination of the two kinds of vision based detection algorithm can effectively reduce the detection error rate and improve the detection accuracy.Parallel information fusion algorithm based on vision and radar,which has better accuracy and lower failure rate than a single sensor vehicle detection algorithm,can effectively overcome the shortcomings of radar’s false and lost detection,and also provide advanced driving assist system with more accurate vehicle detection results at close distance.更多还原