DMSA -- Dense Multi Scan Adjustment for LiDAR Inertial Odometry and Global Optimization

TL;DR

This work introduces Dense Multi Scan Adjustment (DMSA), a dense, feature-free method for registering multiple LiDAR scans by modeling the environment as voxel-based Gaussian landmarks, enabling robust alignment even with small overlaps and dynamic objects. It extends DMSA to LiDAR–IMU odometry through a sliding-window continuous trajectory optimization, adaptive downsampling, and gravity-consistent keyframe representations, achieving strong performance against state-of-the-art baselines on diverse datasets. The method demonstrates improved robustness and accuracy, including seven of eight sequences with the lowest RMSE/max-error, and provides open-source code for reproduction. The approach offers a practical pathway to dense global optimization in LiDAR SLAM and hints at future work in computational efficiency and IMU-free configurations.

Abstract

We propose a new method for fine registering multiple point clouds simultaneously. The approach is characterized by being dense, therefore point clouds are not reduced to pre-selected features in advance. Furthermore, the approach is robust against small overlaps and dynamic objects, since no direct correspondences are assumed between point clouds. Instead, all points are merged into a global point cloud, whose scattering is then iteratively reduced. This is achieved by dividing the global point cloud into uniform grid cells whose contents are subsequently modeled by normal distributions. We show that the proposed approach can be used in a sliding window continuous trajectory optimization combined with IMU measurements to obtain a highly accurate and robust LiDAR inertial odometry estimation. Furthermore, we show that the proposed approach is also suitable for large scale keyframe optimization to increase accuracy. We provide the source code and some experimental data on https://github.com/davidskdds/DMSA_LiDAR_SLAM.git.

Figures (9)

• Figure 1: Resulting sparse keyframe point cloud of sequence exp14 Basement of the Hilti-Oxford Dataset 9968057. The estimated trajectory is marked with a red line and keyframe positions are outlined with red dots.
• Figure 2: Resulting keyframe point cloud from sequence Bicycle Street of our data recording. The scene contains urban areas, field paths and dynamic objects. The trajectory of the sensor platform mounted on a bicycle trailer is marked in red. For areas marked in blue, detail views are provided.
• Figure 3: Exemplary two-dimensional setup with three point clouds. In this case, point cloud 1 (PC1, blue) is used as reference and the relative transformations of point cloud 2 and 3 (PC2 and PC3) to point cloud 1 are optimized. Four voxels contain enough points to be considered as landmark.
• Figure 4: DMSA LiDAR Inertial Odometry overview. Numbers indicate the processing order.
• Figure 5: Outline of the sliding window continuous trajectory optimization. The teal line represents the continuous trajectory within the sliding time window to be optimized. Teal dots symbolize LiDAR points acquired within the time window. Black dots represent Static Points from the map. The points belonging to the landmarks $L_1$ to $L_5$ are marked with black ellipses.