MOBILE LASER SCANNING SYSTEMS FOR GPS/GNSS-DENIED ENVIRONMENT MAPPING

Chenglu Wen, Yifan Xia, Yidu Lian, Yudi Dai, Jin Tan, Chunjin Wang, Jonathan Li2018 · DOI

Abstract

Indoor 3D mapping provides a useful three-dimensional structure via an indoor map for many applications. To acquire highly efficient and relatively accurate mapping for large-scale GPS/GNSS-denied scenes, we present an upgraded backpacked laser scanning system and a car-mounted indoor mobile laser scanning system. The systems provide both 3D laser scanning point clouds and camera images. In this paper, a simultaneous extrinsic calibration approach for multiple multi-beam LIDAR and multiple cameras is also proposed using a Simultaneous Localization and Mapping (SLAM)-based algorithm. The proposed approach uses the SLAM-based algorithm to achieve a large calibration scene using mobile platforms, registers an acquired multi-beam LIDAR point cloud to a terrestrial LIDAR point cloud to acquire denser points for corner feature extraction, and finally achieves simultaneous calibration. With the proposed mapping and calibration algorithms, we can provide centimeter-level colored point clouds with relatively high efficiency and accuracy.

Introduction: The Challenge of Indoor Mapping

Creating accurate 3D maps of indoor spaces is a significant challenge. Unlike outdoor mapping, indoor environments preclude the use of GPS/GNSS for sensor positioning. While robots and handheld scanners exist, they often struggle with scale and "colorization"—the process of overlaying photographic textures onto geometric data.

To produce a "colored" point cloud, the LiDAR sensor must be perfectly aligned (calibrated) with the camera. This is notoriously difficult with mobile multi-beam LiDARs because their point clouds are often too sparse to identify sharp features like corners. Without matching features in both the image and the point cloud, calibration fails, resulting in misaligned 3D models.

Methodology

The researchers developed a hardware-software co-design approach to overcome these limitations.

Hardware Platforms

Two distinct mobile platforms were introduced to handle different scales of indoor environments:

  1. Upgraded Backpack: Features two Velodyne VLP-16 scanners for portability in tight spaces like staircases.
  2. Car-Mounted System: Equipped with a Velodyne HDL-32E, a VLP-16, and four fisheye cameras providing 360-degree visual coverage.
The backpack-mounted (a) and car-mounted (b) mobile laser scanning systems.

SLAM-Based Calibration

To solve the sparsity problem, the authors utilized a "Sparse-to-Dense" calibration technique:

  • Terrestrial Reference: A high-precision Terrestrial Laser Scanner (TLS) scans a calibration scene filled with reflective boards, creating a dense "ground truth" point cloud.
  • SLAM Integration: A feature-based SLAM algorithm builds a preliminary map with the mobile system. This sparse map is then registered to the dense TLS map using an Iterative Closest Point (ICP) algorithm.
  • Extrinsic Calibration: By "densifying" the mobile data with the TLS reference, the corners of reflective boards become identifiable in both the LiDAR data and camera images. Collinearity equations are then used to solve for the exact spatial relationship (extrinsic parameters) between all sensors.
Calibration targets as seen in a camera image (a) and the corresponding LiDAR intensity map (b).

Key Results and Insights

The system was evaluated in diverse real-world environments, including a three-floor building and a massive 200m x 200m underground parking structure.

  • Centimeter-Level Accuracy: The 3D maps achieved high geometric precision, with indoor calibration tests showing a Root Mean Square (RMS) error of approximately 22.5 mm.
  • The TLS Bridge: Using a static, high-precision scanner as an intermediary bypasses the resolution limitations of mobile "puck" style LiDARs.
  • Simultaneous Multi-Sensor Calibration: The SLAM-based approach allows for the simultaneous extrinsic calibration of multiple LiDARs and cameras within a single coordinate system.
  • Fisheye Lens Support: By incorporating the Scaramuzza camera model, the system successfully handles the extreme distortion of fisheye lenses, which are essential for full coverage in confined indoor spaces.

Limitations and Future Outlook

Despite the high accuracy, some challenges remain. Inconsistent lighting in environments like basements can lead to patchy color distribution in the final 3D models. Additionally, while feature detection is largely automated, selecting correspondent corners between images and LiDAR intensity blocks still requires some manual intervention.

Future research will focus on automating the corner-matching process and improving High Dynamic Range (HDR) processing to handle harsh lighting transitions between indoor and outdoor environments.

Summary updated

Abstract

Indoor 3D mapping provides a useful three-dimensional structure via an indoor map for many applications. To acquire highly efficient and relatively accurate mapping for large-scale GPS/GNSS-denied scene, we present an upgraded backpacked laser scanning system and a car-mounted indoor mobile laser scanning system. The systems provide both 3D laser scanning point cloud and camera images. In this paper, a simultaneous extrinsic calibration approach for multiple multi-beam LIDAR and multiple cameras is also proposed using the Simultaneous Localization and Mapping (SLAM)-based algorithm. The proposed approach uses the SLAM-based algorithm to achieve a large calibration scene using mobile platforms, registers an acquired multi-beam LIDAR point cloud to the terrestrial LIDAR point cloud to acquire denser points for corner feature extraction, and finally achieves simultaneous calibration. With the proposed mapping and calibration algorithms, we can provide centimetre-lever coloured point cloud with relatively high efficiency and accuracy.
BibTeX
@inproceedings{wen2018mobile,
  author = {Chenglu Wen and Yifan Xia and Yidu Lian and Yudi Dai and Jin Tan and Chunjin Wang and Jonathan Li},
  title = {{MOBILE LASER SCANNING SYSTEMS FOR GPS/GNSS-DENIED ENVIRONMENT MAPPING}},
  booktitle = {˜The},
  year = {2018},
  doi = {10.5194/isprs-archives-xlii-1-457-2018},
}
Figure