基于特征几何关系的无人车轨迹回环检测

发布时间：2018-01-13 02:37

本文关键词：基于特征几何关系的无人车轨迹回环检测　出处：《中国公路学报》2017年01期 　论文类型：期刊论文

【摘要】：为了在GPS盲区中利用激光雷达构建地图,提出了一种基于回环检测的高精度2D激光点云地图构建方法。首先,从2D激光雷达观测数据中获得无人车的位姿,在2D高斯概率密度空间中提取每帧数据中的环境特征以及求得高斯映射值累加和。其次,利用粒子滤波对车辆位姿与环境特征进行融合优化得到低精度的点云地图和特征地图。然后,利用数据帧中环境特征的数量、车辆位姿以及高斯映射累加和,计算发生轨迹回环的可能性;遍历所有观测帧后得到无人车轨迹回环帧;利用三角剖分法求解回环帧之间的真实转换关系。最后,利用图优化方法得到全局最优的无人车位姿和高精度点云地图与特征地图。试验结果表明:特征地图中同一特征的多次识别结果之间的标准差小于5mm;利用车辆位姿、环境特征和高斯映射累加和能够有效发现路径回环的可能性,其处理2 499帧耗时1.61s;利用Delaunay三角剖分能够准确计算路径回环点,单次运行用时小于1s。
[Abstract]:In order to make use of lidar to construct map in GPS blind area, a high precision 2D laser point cloud map construction method based on loop detection is proposed. The pose of unmanned vehicle is obtained from the observed data of 2D lidar, the environmental feature of each frame data is extracted in 2D Gao Si probability density space, and then the accumulative sum of Gao Si mapping value is obtained. The point cloud map and feature map with low accuracy are obtained by the fusion of vehicle pose and environment feature by particle filter. Then, the number of environmental features in the data frame, vehicle pose and Gao Si map are added together. To calculate the possibility of the return loop of the locus; After traversing all the observation frames, the loop frame of the track of the unmanned vehicle is obtained. Triangulation method is used to solve the real conversion relationship between the return frames. Finally. By using the method of map optimization, the global optimal unmanned vehicle posture and high accuracy point cloud map and feature map are obtained. The experimental results show that the standard deviation between the multiple recognition results of the same feature in the feature map is less than 5 mm. By using vehicle posture, environmental features and Gao Si mapping, the possibility of effectively discovering the path return loop can be found, and the processing time of 2 499 frames is 1.61s; By using Delaunay triangulation, the loop point of the path can be calculated accurately, and the time used in a single operation is less than 1 s.
【作者单位】：长安大学信息工程学院;
【基金】：教育部高等学校学科创新引智计划项目(B14043)
【分类号】：U495;U463.6
【正文快照】： 0引言无人车的自主行驶必须通过“自主定位”确定自身当前位置,通过“路径规划”在地图中选择到达目的地的最优行驶路线,通过“行为决策”动态地调整局部行驶路线,实现对交通事件的实时处理。在正常道路交通环境中,无人车利用全球导航卫星系统构建的路网文件实现路径规划,并在

【参考文献】