Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Geo-Registration of Terrestrial LiDAR Point Clouds with Satellite Images without GNSS

Created by
  • Haebom

Author

Xinyu Wang, Muhammad Ibrahim, Haitian Wang, Atif Mansoor, Ajmal Mian

Outline

This paper presents a novel method to solve the difficulty of accurate georegistration of LiDAR point clouds in urban areas with many tall buildings and bridges where GNSS signals are blocked. Existing methods rely on real-time GNSS and IMU data, require pre-calibration, and assume stable position estimation during data collection, which is not true in urban areas. In this paper, we propose a structural georegistration and spatial correction method that aligns satellite images and 3D point clouds without pre-positioning information, enabling frame-by-frame GNSS information recovery and city-scale 3D map generation. We segment road points using the Point Transformer model, and extract and align road skeletons and intersections from the point clouds and target map. We perform global rigid body alignment using the intersections, and perform local fine-tuning using RBF interpolation. We apply elevation correction based on the topographic information of the SRTM dataset to resolve vertical mismatch. Experimental results on the KITTI benchmark and Perth CBD datasets show that the average planar alignment standard deviation is improved to 0.84 m (55.3% improvement) on the KITTI dataset and to 0.96 m (77.4% improvement) on the Perth dataset, while the elevation correlation is also improved.

Takeaways, Limitations

•
Takeaways:
◦
We demonstrate that accurate LiDAR point cloud georegistration is possible even in urban environments without GNSS signals.
◦
Overcoming the dependence of prior position information on Limitations of existing methods.
◦
Demonstrates the potential for creating city-scale 3D maps using satellite imagery.
◦
Improving accuracy through road feature extraction based on Point Transformer.
◦
Performance verification through experiments on KITTI and Perth datasets.
•
Limitations:
◦
It may perform better on certain types of roads (e.g. roads with intersections). Generalization to other types of terrain is needed.
◦
Elevation correction performance may be affected by the accuracy of SRTM data. It is necessary to find ways to utilize more accurate elevation information sources.
◦
The bias of the data used to train the Point Transformer model can affect the results. It is necessary to train the model using various datasets.
◦
The amount of calculation can be relatively large. Algorithm optimization for real-time processing is required.
👍
Made with Slashpage