基于多帧叠加激光点云的可通行区域提取研究
发布时间：2018-10-12 09:22
【摘要】：无人驾驶中的环境感知系统是实现无人驾驶的重要部分,也是无人驾驶车辆安全行驶的第一环,而可通行区域则是环境感知系统中的基础工作。基于Velodyne HDL-32E激光雷达,本文提出了基于多帧激光点云数据的可通行区域提取方法和框架。具体研究内容包含以下几个方面:1)在介绍Velodyne HDL-32E激光雷达特性和数据结构的基础上,提出了在多帧点云数据中进行可通行区域提取的框架:首先在单帧激光点云数据中提取可通行区域,然后基于POS数据或者SLAM算法对连续多帧数据进行对准,最后以单帧数据中的提取结果为基础,在对准的点云数据中再次进行可通行区域提取,得到最终结果。2)研究单帧点云数据中可通行区域的提取方法。提出基于竖直平面上连续点夹角的可通行区域提取方法,确定了对于不同激光扫描线角度阈值的计算方法。3)对比研究了基于POS和基于SLAM算法的连续多帧点云数据对准方法。针对可通行区域提取这一激光点云应用实例,提出了基于主成份分析的点云对准效果评价方法。4)研究了在对准后多帧激光雷达点云中实时提取可通行区域的方法。首先基于体素(Voxel)的方法对障碍物点云中的重复、邻近点进行了去除。然后基于八叉树在对准后的原始点云中建立空间索引。最后通过邻近点的高度差对障碍物点进行验证,剔除障碍物点云中存在的误检点。通过使用多个真实场景采集的数据的实验验证,证明了本文可通行区域提取方法可同时运用于城市环境和野外环境。
[Abstract]:The environment sensing system in driverless is an important part of realizing unmanned driving and the first link of driving safety of driverless vehicle. The passable area is the basic work in the environment sensing system. Based on Velodyne HDL-32E lidar, this paper presents a method and framework for extracting passable region based on multi-frame laser point cloud data. The specific research contents include the following aspects: 1) on the basis of introducing the characteristics and data structure of Velodyne HDL-32E lidar, A framework for extracting passable regions from multi-frame point cloud data is proposed. Firstly, the passable region is extracted from single-frame laser point cloud data, and then the continuous multi-frame data is aligned based on POS data or SLAM algorithm. Finally, based on the extraction results of single frame data, the passable region is extracted again in the aligned point cloud data, and the final results are obtained. 2) the method of extracting passable area from single frame point cloud data is studied. In this paper, a method of extracting passable region based on the angle of continuous points on vertical plane is proposed, and the calculation method of angle threshold of different laser scanning lines is determined. 3) the method of continuous multi-frame point cloud data alignment based on POS and SLAM algorithm is compared and studied. Aiming at the application example of laser point cloud extraction from passable area, a point cloud alignment evaluation method based on principal component analysis (PCA) is proposed. 4) the method of extracting passable area in multi-frame lidar point cloud after alignment is studied in real time. Firstly, the repeat and adjacent points in obstacle point cloud are removed based on voxel (Voxel) method. Then the spatial index is built based on octree in the original point cloud after alignment. Finally, the obstacle points are verified by the height difference of the adjacent points, and the false detection points in the obstacle point cloud are eliminated. It is proved that this method can be used in both urban environment and field environment through the experimental verification of data collected from many real scenes.
【学位授予单位】：武汉大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN958.98
本文编号：2265599
[Abstract]:The environment sensing system in driverless is an important part of realizing unmanned driving and the first link of driving safety of driverless vehicle. The passable area is the basic work in the environment sensing system. Based on Velodyne HDL-32E lidar, this paper presents a method and framework for extracting passable region based on multi-frame laser point cloud data. The specific research contents include the following aspects: 1) on the basis of introducing the characteristics and data structure of Velodyne HDL-32E lidar, A framework for extracting passable regions from multi-frame point cloud data is proposed. Firstly, the passable region is extracted from single-frame laser point cloud data, and then the continuous multi-frame data is aligned based on POS data or SLAM algorithm. Finally, based on the extraction results of single frame data, the passable region is extracted again in the aligned point cloud data, and the final results are obtained. 2) the method of extracting passable area from single frame point cloud data is studied. In this paper, a method of extracting passable region based on the angle of continuous points on vertical plane is proposed, and the calculation method of angle threshold of different laser scanning lines is determined. 3) the method of continuous multi-frame point cloud data alignment based on POS and SLAM algorithm is compared and studied. Aiming at the application example of laser point cloud extraction from passable area, a point cloud alignment evaluation method based on principal component analysis (PCA) is proposed. 4) the method of extracting passable area in multi-frame lidar point cloud after alignment is studied in real time. Firstly, the repeat and adjacent points in obstacle point cloud are removed based on voxel (Voxel) method. Then the spatial index is built based on octree in the original point cloud after alignment. Finally, the obstacle points are verified by the height difference of the adjacent points, and the false detection points in the obstacle point cloud are eliminated. It is proved that this method can be used in both urban environment and field environment through the experimental verification of data collected from many real scenes.
【学位授予单位】：武汉大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN958.98
【参考文献】
相关期刊论文 前10条
1 付黎明;李必军;叶雨同;;基于多尺度重采样的车道线检测[J];电子技术应用;2017年04期
2 张永博;李必军;陈诚;;激光点云在无人驾驶路径检测中的应用[J];测绘通报;2016年11期
3 叶茂胜;李必军;莫柠锴;孔格菲;;基于色域差分与伽马校正的交通灯识别[J];软件导刊;2016年09期
4 黄如林;梁华为;陈佳佳;赵盼;杜明博;;基于激光雷达的无人驾驶汽车动态障碍物检测、跟踪与识别方法[J];机器人;2016年04期
5 潘安;李必军;;基于Android智能手机的前方车辆检测系统[J];软件导刊;2016年05期
6 马飞跃;王晓年;;无人驾驶汽车环境感知与导航定位技术应用综述[J];汽车电器;2015年02期
7 ;2014年中国智能车未来挑战赛在江苏常熟举办[J];工业控制计算机;2014年12期
8 李飞;栾学晨;杨必胜;李秋萍;;高等级道路网的拓扑自动保持方法[J];武汉大学学报(信息科学版);2014年06期
9 栾学晨;范红超;杨必胜;李秋萍;;城市道路网主干道提取的形态分析方法[J];武汉大学学报(信息科学版);2014年03期
10 杨必胜;董震;魏征;方莉娜;李汉武;;从车载激光扫描数据中提取复杂建筑物立面的方法[J];测绘学报;2013年03期
，本文编号：2265599
本文链接：https://www.wllwen.com/kejilunwen/xinxigongchenglunwen/2265599.html
