LiDAR, GNSS and IMU Sensor Alignment through Dynamic Time Warping to Construct 3D City Maps

Haitian Wang, Hezam Albaqami, Xinyu Wang, Muhammad Ibrahim, Zainy M. Malakan, Abdullah M. Algamdi, Mohammed H. Alghamdi, Ajmal MianarXiv 2025 · DOI · arXiv

Introduction

High-fidelity 3D city mapping is a cornerstone of autonomous navigation and urban planning. However, generating these maps in "urban canyons"—dense metropolitan areas with tall buildings—presents significant challenges. In such environments, GNSS signals are often blocked or reflected, leading to positioning errors. Furthermore, hardware-level synchronization between LiDAR, IMU, and GNSS sensors is frequently plagued by clock drift and transmission latency, which causes "ghosting" and misalignments in the final 3D reconstruction.

Methodology: The LIGMA Framework

To address these issues, we present the LiDAR-IMU-GNSS Mapping Architecture (LIGMA). Unlike traditional methods that rely on rigid timestamp matching, LIGMA employs a data-driven "soft synchronization" approach.

Soft Synchronization via Dynamic Time Warping (DTW)

The core innovation of LIGMA is using velocity as a universal language to align disparate sensor modalities. The system independently calculates velocity profiles for the LiDAR (via scan matching) and the IMU/GNSS (via an Extended Kalman Filter). Dynamic Time Warping (DTW) is then applied to find the optimal temporal alignment between these velocity signatures, effectively bypassing the inaccuracies of hardware timestamps. To handle potential ambiguities in DTW during constant-speed segments, the authors implemented "stationary anchors"—forcing matches during vehicle stops—and restricted the warping window using a Sakoe–Chiba band.

The LIGMA soft synchronization process aligns sensor timelines by matching data-driven velocity profiles.

Hierarchical Registration and Anchoring

The mapping pipeline follows a hierarchical structure:

  1. Local Registration: Sequential frames are aligned using the Normal Distributions Transform (NDT).
  2. Global Optimization: A Pose Graph Optimization (PGO) framework, incorporating loop closure detection via Scan Context, ensures long-term consistency.
  3. Spatial Anchoring: High-quality RTK-GNSS signals serve as spatial anchors to tie the local map to real-world coordinates, with weights adjusted based on signal quality.
  4. Fine Merging: Overlapping segments are refined using the Iterative Closest Point (ICP) algorithm to ensure seamless transitions.

Experimental Setup

The framework was validated using a custom multimodal dataset captured in the Perth Central Business District (CBD), Western Australia. The dataset covers 18.6 km of urban roads across 21 loops, totaling 144,000 frames.

The mobile mapping platform equipped with an Ouster OS1 LiDAR and a RUG-3 GNSS/IMU system.

Results and Evaluation

LIGMA was evaluated against standard LiDAR-only (NDT-ICP) baselines using a dual-metric approach that measures both global trajectory accuracy and local geometric fidelity. Global accuracy was assessed by comparing extracted road centerlines to cartographic ground truth, while local fidelity was measured via intersection centroid offsets.

  • Global Alignment: The framework reduced the average road centerline Root Mean Square Error (RMSE) from 3.32 m to 1.24 m, representing a 61.4% improvement.
  • Local Fidelity: Intersection centroid offsets were reduced by 84.8%, dropping from 13.22 m to 2.01 m.
  • Efficiency: The system processed the 18.6 km trajectory in under 15 hours on a standard workstation, demonstrating its viability for large-scale offline mapping.

Conclusion

The LIGMA framework demonstrates that data-driven temporal synchronization via DTW significantly outperforms traditional timestamp-based alignment in complex urban environments. By treating velocity as a shared physical reality across sensors, the system achieves high-fidelity 3D maps even in the presence of GNSS degradation and sensor jitter. Future work will explore real-time adaptations and the integration of semantic features to further enhance mapping robustness.

Summary updated

BibTeX
@misc{wang2025lidar,
  author = {Haitian Wang and Hezam Albaqami and Xinyu Wang and Muhammad Ibrahim and Zainy M. Malakan and Abdullah M. Algamdi and Mohammed H. Alghamdi and Ajmal Mian},
  title = {{LiDAR, GNSS and IMU Sensor Alignment through Dynamic Time Warping to Construct 3D City Maps}},
  year = {2025},
  eprint = {2507.08420},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2507.08420},
}
Figure