Towards Effective LiDAR-Based Mapping with Handheld Rotating Sensors in Complex Environments

초록

Mobile LiDAR systems facilitate 3D mapping in complex environments featuring varied terrain and structures. To address the limited field of view of compact LiDAR sensors, mechanical rotation can be employed to extend spatial coverage. However, this method results in sparse point distributions and decreased frame-to-frame overlap, presenting significant challenges for conventional SLAM algorithms in achieving reliable scan matching and consistent mapping. This study investigates SLAM performance under these conditions through experiments using a manually rotated Velodyne VLP-16 sensor. Data were collected along a trajectory combining outdoor building perimeters and indoor corridors, relying solely on LiDAR without GPS, IMU, or camera assistance. Preliminary tests with HDL-Graph-SLAM diverged under the sparse, low-overlap conditions, whereas KISS-ICP provided stable odometry with residual drift. Building on this, KISS-SLAM was analyzed in detail by tuning parameters such as max_range, voxel_size, and local map resolution. Results show that reducing max_range to 20 m improved odometry stability in narrow indoor spaces and adjusting voxel_size to approximately 0.01 × max_range yielded generally consistent performance. Loop closure detection was highly sensitive to local map resolution: default settings failed to recognize revisits, whereas tuned parameters substantially increased loop closure detections, effectively reducing drift and improving global consistency. While finer local map resolution improves mapping accuracy, it also increases computational cost, underscoring the importance of selecting parameters that balance accuracy with efficiency. These findings highlight adaptive strategies for handling sparse and irregularly sampled point clouds, demonstrating that compact rotating LiDAR systems are feasible for LiDAR-only mapping in GPS-denied environments, as long as parameter settings are carefully tuned to balance accuracy with computational efficiency. © ACRS 2025.All rights reserved.

키워드