基于点云配准的多固态激光雷达标定算法

Calibration Algorithm of Multi Solid-state Lidar Based on Pointcloud Registration

摘要:

针对多个固态激光雷达协同工作时,需要准确地进行外参标定的实际需求,提出一种基于点云配准的多固态激光雷达自动标定算法。该标定算法由标定物分割、初始配准和精确配准三个阶段构成。在标定物分割阶段,首先通过叠加多帧非重复扫描数据制作标定点云,再使用半径滤波和体素下采样滤波分割出纹理特征明显的目标点云。在初始配准阶段,使用3D-HARRIS算法提取关键点,并使用方向直方图(SHOT)特征描述子进行特征描述,然后匹配对应点并使用采样一致算法完成初始配准；在精配准阶段使用迭代最近邻(ICP)算法进行精确配准,从而获得精确的外参标定效果。在Bunny兔数据集和现场获得的数据上进行实验,结果表明,当保证配准平均误差小于1mm的前提下,所提出算法的性能优于多种现有算法。

Abstract:

Aiming at the practical requirement of accurately calibrating external parameters of multiple solid-state lidars, a multi-solid-state lidar automatic calibration algorithm based on point cloud registration is proposed. The calibration algorithm consists of three stages:calibration object segmentation, initial registration and precise registration. In the target segmentation stage, a calibration point cloud is made by superimposing multiple frames of non-repetitive scan data, and then radius filtering and voxel down-sampling filtering are used to segment the target point cloud with obvious texture characteristics. In the initial registration stage, key points are extracted by 3D-HARRIS algorithm, and then key points are described by the signature of histogram of orientation (SHOT) feature descriptor, then the sampling consensus algorithm is performed to calculate the corresponding points and complete the initial registration. In the fine registration stage, the iterative closest point (ICP) algorithm performs precise registration to obtain precise external parameter calibration effects. Experiments are conducted on the Bunny rabbit dataset and the data obtained on site. The results show that the performance of the proposed algorithm is better than many existing algorithms under the premise that the average registration error is less than 1mm.