Efficient Multi-floor Indoor Mapping with A Versatile Rotating LiDAR System

Yangzi Cong, Tianhe Xu, Chi Chen, Wenfeng Nie, Bisheng Yang2024 · DOI

This paper, presented at the 2024 ISPRS TC IV Mid-term Symposium, introduces a robust solution for one of the trickiest problems in mobile robotics: creating accurate 3D maps of complex, multi-story indoor environments. By mounting a standard LiDAR on a rotating platform and tightly integrating its data with an IMU and motor feedback, the researchers have developed a system that overcomes the field-of-view limitations that typically plague indoor mapping.

TL;DR

The researchers developed a rotating LiDAR-inertial system that uses an iterative Kalman Filter and sliding-window optimization to map multi-floor buildings. By rotating the sensor, they significantly increase the field of view, reducing "blind spots" and improving mapping consistency in feature-poor environments like stairwells and long hallways.

What Problem Does This Paper Address?

Standard Multi-Beam LiDAR (MBL) sensors often have a very narrow vertical field of view (FoV) and low vertical resolution compared to their horizontal performance. In indoor environments—especially multi-story buildings with narrow hallways and staircases—this leads to "sensing degeneration." Essentially, the sensor doesn't see enough unique geometry to accurately determine its position, leading to significant "drift," particularly in the Z-axis (height).

While rotating a LiDAR can solve the FoV problem, it introduces a new challenge: high dynamic motion. The sensor is moving so fast that the resulting point cloud is heavily distorted (skewed). Correcting this distortion while maintaining "global consistency" (ensuring the map doesn't warp as you move between floors) is the core challenge the authors tackle.

Methods

The system uses a Velodyne 16-beam LiDAR mounted on an electric motor, paired with an Xsens IMU. The technical "secret sauce" lies in a two-stage software pipeline:

  1. The Front-End (Odometry): They use an iterative Error State Kalman Filter (ESKF) to tightly integrate measurements from the motor, IMU, and LiDAR. By knowing the exact angle of the motor at any given millisecond, they can "de-skew" the laser points in real-time to account for the rotation.
  2. The Back-End (Mapping): To ensure the map stays aligned over long distances and multiple floors, they employ a sliding-window-based batch optimization (based on the BALM framework).
  3. Adaptive Voxelization: The system doesn't just dump points into a grid. It uses an octree structure to voxelize the map at adaptive resolutions. It then performs Principal Component Analysis (PCA) within these voxels to identify and extract linear (edges) and planar (walls/floors) features, which are much more reliable for alignment than raw points.

Datasets

The authors evaluated their system using:

  • Real-world Indoor Data: Collected over 5 minutes in a research lab building featuring a mix of wide and narrow multi-floor hallways.
  • Ground Truth: They compared their results against a high-precision map generated by a Leica VZ400 Terrestrial Laser Scanner (TLS), which has a measuring accuracy of 3mm.
  • SOTA Comparison: They benchmarked their performance against FAST-LIO2, a current state-of-the-art LiDAR-inertial odometry method.

Key Insights and Ideas

  • The 90°/s "Sweet Spot": The researchers tested various rotation speeds and found that 90°/s provided the best balance. At this speed, the walls in the resulting map were at their thinnest, indicating the highest precision and lowest noise.
  • Rotation as a Stability Tool: Beyond just seeing more of the room, the rotation mechanism acts as a safeguard against "degeneracy." In a long, smooth hallway, a static LiDAR might lose track of its forward motion. A rotating LiDAR, however, is constantly catching glimpses of the ceiling, floor, and distant doorways at varying angles, providing more constraints for the localization algorithm.
  • Hardware-Level Sync: The authors emphasized that the clock of every sensor is hard-synchronized through an ARM processor. In high-speed rotating systems, even a few milliseconds of lag between the motor angle and the laser pulse can ruin the map.

Key Results

  • Reduced "Holes": Qualitative results show that the rotating mechanism successfully filled in large gaps in the point cloud that appeared when the LiDAR was kept stationary.
  • Superior Multi-floor Consistency: When compared to FAST-LIO2, the proposed method showed significantly less deviation. FAST-LIO2 exhibited noticeable "distortions" and misalignments between different floors, whereas the authors' batch optimization rectified these errors.
  • Precision: The system achieved "thinner" wall representations in the point cloud, suggesting a tighter fit to the actual environment geometry compared to standard setups.

Limitations and Future Work

While the system is effective, the authors noted a few areas for improvement:

  • Degenerative Scenes: Even with rotation, some "slippage" occurred in extremely featureless environments.
  • Optimization Stability: The stability of the batch optimization back-end needs further stress-testing in more diverse and "messy" real-world scenarios.
  • Dataset Expansion: The current study relied on a single building; the authors plan to include more varied datasets in the future to prove the system's versatility across different architectural styles.

Summary updated

Abstract

Light Detection And Ranging (LiDAR) technology has provided an impactful way to capture 3D environmental map. However, consistent mapping in sensing-degenerated and perceptually-limited scenes (e.g. multi-story buildings) or under high dynamic sensor motion (e.g. rotating platform) remains a significant challenge. To this end, an efficient multi-floor indoor mapping system is proposed in this paper utilizing a versatile rotating LiDAR platform. In the front end, measurements from motor, IMU and LiDAR are tightly integrated to track the fast motion state of system, based on iterative Error State Kalman Filter (ESKF). Then linear and planar features are extracted from the point cloud map voxelized with adaptive resolutions. A sliding-window-based batch optimization is performed to simultaneously optimize the states of local frames with reference to the map consistency. In the experiments, we investigated the influence of rotating speed on the mapping performance as well as the superiority of rotating mechanism when compared to standard LiDAR setup. Moreover, comparative studies with one of the SOTA work, FAST-LIO2, have shown the competitive mapping results in multi-floor indoor environments.
BibTeX
@inproceedings{cong2024efficient,
  author = {Yangzi Cong and Tianhe Xu and Chi Chen and Wenfeng Nie and Bisheng Yang},
  title = {{Efficient Multi-floor Indoor Mapping with A Versatile Rotating LiDAR System}},
  booktitle = {˜The},
  year = {2024},
  doi = {10.5194/isprs-archives-xlviii-4-2024-613-2024},
}
Figure