论文部分内容阅读
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。
In order to construct a map by using lidar in the blind area of GPS, a method of building a 2D laser point cloud map based on loop detection is proposed. Firstly, the pose of UAV is obtained from the 2D lidar observation data, the environmental features in each frame of data are extracted in 2D Gaussian probability density space, and the cumulative sums of Gaussian map values are obtained. Secondly, the particle filter is used to optimize the vehicle pose and environment features to obtain a low-precision point cloud map and feature map. Then, using the number of environmental features in the data frame, the pose of the vehicle and the cumulative sum of the Gaussian maps, the possibility of trajectory looping is calculated. After all the observation frames are traversed, unmanned vehicle trajectory loop frames are obtained. The triangulation method is used to solve the loopback frames Between the true conversion relations. Finally, the graph optimization method is used to get the global optimal unmanned vehicle pose and high-precision point cloud map and feature map. The test results show that the standard deviation between multiple recognition results of the same feature in the feature map is less than 5 mm. Using the vehicle pose, environment feature and Gaussian map cumulative sum, the probability of loop loop can be effectively found. 1.61s; Delaunay triangulation can accurately calculate the loop back to the point, a single run with less than 1s.