中文摘要：

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。

英文摘要：

In order to creating map based on lidar in GPS-denied environments,a high precision2 Dlaser point-clouds map creating method based on loop closure detection was proposed.Firstly,the poses of unmanned vehicle were extracted from the observation data of 2Dlidar,and the environment features and the sum of Gaussian mapping value of frame data in the bivariate Gaussian probability distribution space were extracted.Secondly,the low precision point-clouds map and feature map were created by fusion optimization of poses and features based on particle filter.Then the amount of environment feature of data frame,pose of unmanned vehicle,and the sum of Gaussian mapping were used to compute the trajectory loop closure likelihood,and the trajectory loop closure frames were obtained after going through all the observation frames,and the Delaunay triangulation was used to compute the real transformation among loop closure frames.Finally,agraph optimization-based method was utilized to get a global optimization solution to the vehicle poses,the higher precision point-clouds map and the features map.The results show that the standard deviation between the multiple recognition results from the samefeature is less than 5mm in feature map.The vehicle poses,environment features and the sum of Gaussian mapping could effectively explore the loop closure likelihood of trajectory,and the elapsed time in processing 2 499 frames is 1.67 s.The Delaunay triangulation could accurately compute the points of loop closure,and the once uptime is less than 1s.