What's in My LiDAR Odometry Toolbox?

Abstract

With the democratization of 3D LiDAR sensors, precise LiDAR odometries and SLAM are in high demand. New methods regularly appear, proposing solutions ranging from small variations in classical algorithms to radically new paradigms based on deep learning. Yet it is often difficult to compare these methods, notably due to the few datasets on which the methods can be evaluated and compared. Furthermore, their weaknesses are rarely examined, often letting the user discover the hard way whether a method would be appropriate for a use case.In this paper, we review and organize the main 3D Li-DAR odometries into distinct categories. We implemented several approaches (geometric based, deep learning based, and hybrid methods) to conduct an in-depth analysis of their strengths and weaknesses on multiple datasets, guiding the reader through the different LiDAR odometries available. Implementation of the methods has been made publicly available at https://gitlab.kitware. com/keu-computervision/pylidar-slam.

CT-ICP: Real-time Elastic LiDAR Odometry with Loop Closure

“…This is the case for NCLT [33], a dataset containing 27 long sequences (≥ 20000 scans per sequence) of scans acquired at Michigan University using a Velodyne HDL32 mounted on a two-wheeled Segway. This dataset is particularly challenging, because the vehicule introduces abrupt rotations to the LiDAR's own rotation axis, which is problematic for classical ICP-based odometry methods, as shown in [34] NCD [35], the Newer College Dataset, contains two sequences (∼15000 and ∼26000 scans) of a handheld Ouster 64-channel LiDAR mounted on a stick, that is carried accross the Oxford campus.…”

“…For quantitative evaluations, as in [11], [15], [4], [34], we use the KITTI Relative Translation Error (RTE), which averages the trajectory drift over segments of lengths ranging from 100 m to 800 m. When computing the score on multiple sequences, the average (AVG) is computed over all the segments of all sequences (different from the mean of RTE over the sequences), and mirrors KITTI's benchmark evaluation.…”

“…To compare our method, we evaluated three other LiDAR odometries on the datasets presented: MULLS [4], IMLS-SLAM [15], and pyLIDAR-SLAM F2M [34]. For each of these methods, we tested a whole set of parameters to get the best results, but once the parameters are found, we kept the same for all the driving datasets.…”

“…For each of these methods, we tested a whole set of parameters to get the best results, but once the parameters are found, we kept the same for all the driving datasets. MULLS and IMLS-SLAM are odometries specialized for driving datasets, which is why we only compared our odometry to pyLIDAR-SLAM F2M [34] on the high-frequency motion datasets NCD and NCLT.…”