LiLoc: LiDAR Image Localization

Scripts for localizing image sets in point clouds. Based on XFeat and LighterGlue.

Installation

Clone the repository including submodules.

git clone https://github.com/RUB-Informatik-im-Bauwesen/LiLoc --recurse-submodules

Install the requirements. For GPU support use the installation instructions from pytorch.org.

pip install torch torchvision torchaudio  # Replace with GPU versions from pytorch.org
pip install -r accelerated_features/requirements.txt
pip install -r requirements.txt

Usage

Quickstart: Matches all images in a given folder and writes to ./example/defect_img/matches

python liloc.py match ./example/defect_img/

To match all images from one folder against all images from another folder (e.g. camera photos against LiDAR photos).

python liloc.py cross_match ./example/scan_img/ ./example/defect_img/

For more flags and information use the help function:

python liloc.py match --help
python liloc.py cross_match --help

---

LiLoc can also extract rectified images from point clouds.

python extract_images_from_e57.py pointcloud.e57

If there are no images, use the --rgb flag to render rectified images from the point cloud's rgb values.

Paper

To be published at ISARC 2025 in Montréal, Canada

Acknowledgement

This repository is based on parts of the research project carried out at the request of the BMDV, represented by the Federal Highway Research Institute, under research project No. 69.0017/2023. The developers are solely responsible for the content.