基于双多线激光雷达的道路环境感知算法研究与实现

发布时间：2018-03-05 02:36

本文选题：环境感知　切入点：激光雷达数据融合　出处：《南京理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：无人车在军事与民用方面具有广泛的应用前景。随着物联网、人工智能、计算机科学等相关技术的发展,无人车的外在环境也日臻完善。环境感知作为无人车系统的重要组成部分,对整个车起着至关重要的作用。本文针对无人车环境感知中的两个重点与难点问题进行研究,设计了基于双激光雷达的环境感知处理架构,并基于该架构研究与实现了结构化环境下的低矮道边检测与非结构化环境下的负障碍检测两个环境感知的难点问题。本文的主要研究成果与创新点如下:1、针对以往单个单线、多线激光雷达点云密度小,对于特殊场景感知能力差的特点,研究设计了基于双多线激光雷达对称式安装的环境感知与信息融合处理架构,并对该架构下的点云密度进行定量分析。实验表明,该架构大大提高了无人车车体前方观测区的点云密度,减小了车体周身盲区,可以解决无人车难度较大的环境感知问题。2、根据双激光雷达水平安装的雷达扫描特点,分析了雷达点在障碍区的分布特性,提出一种新的结构化环境下的低矮道边的检测算法。算法使用了基于梯度一致性的点云分割方法,该方法可对雷达点进行快速分割,高效的将雷达点分割为地面点与障碍物点。然后利用路面点与栅格地图提取出候选道边点,最后分别使用最小二乘与改进的RANSAC算法进行道边提取。实验结果显示,点云分割算法具有良好的分割效果,改进后的RANSAC算法具有较高的实时性,能够满足无人车的需求。3、针对非结构化环境下的负障碍检测问题提出一种新的感知方法,该方法不依赖于地面平整度,通过局部点云分布特征进行检测。首先,将雷达点云映射到多尺度栅格,统计各栅格的点云密度与相对高度等特征并做标记;然后,从点云数据中抽取负障碍几何特征,将栅格的统计特征与负障碍的几何特征进行多特征关联找到关键特征点对;最后,将特征点对聚类,划分负障碍。方法已成功运行在无人车上,实验表明,该方法具有较高的实时性和可靠性,在非结构化环境下具有良好的检测效果。上述研究成果均已成功使用在"行健一号"无人车上,该车多次参加"中国智能车未来挑战赛",并在比赛中取得优异的成绩。
[Abstract]:With the development of Internet of things, artificial intelligence, computer science and other related technologies, As an important part of the unmanned vehicle system, environmental awareness plays an important role in the whole vehicle. The environment sensing processing architecture based on dual lidar is designed. Based on this architecture, the paper studies and realizes two difficult problems of environmental perception, which are low lane edge detection in structured environment and negative obstacle detection in unstructured environment. The main research results and innovations in this paper are as follows: 1, aiming at single line in the past. Because of the low point cloud density of multi-line lidar and the poor perceptual ability of special scene, the environment sensing and information fusion processing architecture based on symmetrical installation of dual-line lidar is studied and designed. The experimental results show that the point cloud density of the observation area in front of the vehicle body is greatly increased, and the blind area around the body is reduced. It can solve the problem of environment perception, which is difficult for unmanned vehicle. According to the radar scanning characteristics installed horizontally with double lidar, the distribution characteristics of radar points in obstacle area are analyzed. In this paper, a new algorithm for detection of low edge in structured environment is proposed. The algorithm uses a point cloud segmentation method based on gradient consistency, which can segment radar points quickly. The radar points are divided into ground points and obstacle points efficiently. Then the candidate edge points are extracted from the road surface points and raster maps. Finally, the least square algorithm and the improved RANSAC algorithm are used to extract the edge points. The experimental results show that, The point cloud segmentation algorithm has a good segmentation effect, and the improved RANSAC algorithm has a high real-time performance, which can meet the requirements of unmanned vehicles. A new perception method is proposed for the negative obstacle detection problem in unstructured environment. This method does not depend on the smoothness of the ground, and detects the distribution characteristics of the local point cloud. Firstly, the radar point cloud is mapped to the multi-scale grid, and the point cloud density and the relative height of each grid are counted and marked. The geometric features of negative obstacle are extracted from point cloud data, and the statistical features of grid and geometric features of negative obstacle are correlated to find the key feature points. Finally, the feature points are clustered. The method has been successfully run on an unmanned vehicle. Experiments show that the method has high real-time and reliability. The above research results have been successfully used in the "Xingjian 1" unmanned vehicle, which has participated in the "China Smart vehicle Future Challenge" many times, and has achieved excellent results in the competition.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U463.6;TN958.98

【参考文献】