A Multi-Feature Base Multi-Adaptation Lidar Odometry

To address the problems that the current feature-based lidar odometry algorithms have difficulties in achieving high accuracy based on sparse features and adapting to different kinds of lidar at the same time, a multi-adaptation lidar odometry algorithm based on multi-features (MFMA-LOAM) is proposed. In addition to the traditional edge points and plane points, intensity points and ground points are extracted as supplementary features. Among them, the intensity point features that meet the requirements are extracted based on lidar intensity reflection information. The ground feature extraction method is improved by using the estimated normal vector to achieve simultaneous adaptation of multiple types of lidar. Further, in order to ensure acceptable speed and improve the alignment accuracy at the same time, a two-stage alignment method is applied to complete the positional calculation, in which feature weights are added to improve the alignment accuracy and categorical nearest neighbor search is used to speed up the alignment. Finally, experimental validation is carried out using the KITTI dataset and the self-collected dataset respectively; the experimental results show that the MFMA-LOAM algorithm with satisfactory real-time performance is more accurate than related classical feature-based algorithms and can be adapted to both solid-state lidar and mechanical lidar at the same time.