Automatic Registration Estimation Based on ArUco Markers

About

This project automatically computes the correct scale and registration of 3D reconstructions generated with COLMAP by placing one or more ArUco markers in the scene. This module does not perform the reconstruction task itself so a colmap installation isn’t needed.

Installation

This repository is tested on Python 3.6+ and can be installed from PyPI:

pip install aruco-estimator

pip install git+https://github.com/meyerls/aruco-estimator

Usage

Test Data

Download and extract the test dataset:

wget -O door.zip "https://faubox.rrze.uni-erlangen.de/dl/fiUNWMmsaEAavXHfjqxfyXU9/door.zip"
unzip door.zip

Registration

Test the registration functionality with the example project:

aruco-estimator register ./door --target-id 7 --dict-type 4 --show --aruco-size 0.15

Merging

Scripting

Test the registration functionality with the example project:

aruco-estimator align .data/_park .data/_alley--target-id 7 --dict-type 4 --show --aruco-size 0.15

from aruco_estimator.sfm.colmap import COLMAPProject
from aruco_estimator.utils import get_transformation_between_clouds,get_corners_at_origin
import cv2
project = COLMAPProject('./door', )
/sparse/0
target_id = 7
aruco_size = .15
aruco_results = project.detect_markers(dict_type=cv2.aruco.DICT_4X4_50)

# Get 3D corners for normalization
target_corners_3d = aruco_results[target_id]
print(target_corners_3d) 

# Calculate 4x4 transform with scaling so tag is at the origin 
transform = get_transformation_between_clouds(target_corners_3d, get_corners_at_origin(side_length=aruco_size))

# Apply normalization to the project
print("Normalizing poses and 3D points...")
project.transform(transform)
project.save("./transformed_output/")

Dense Clouds

Current Script expects {project_dir}/fused.ply for dense cloud alignment and viz

Known Limitations

Only SIMPLE_RADIAL and PINHOLE camera models are supported
Aruco boards are not uniquely supported
Pose estimation is not robust to false detections; ransac would be beneficial
Only COLMAP .bin and .txt models are supported

Roadmap

Update README with multi tag examples
Improved pose estimation robustness
Implement the merge by tag tool
Support for additional camera models
Geo-referencing of ArUco markers with Earth coordinate system using GPS or RT

Troubleshooting

OpenCV ArUco Module Issues

If cv2 doesn’t detect the ArUco marker module, try reinstalling OpenCV:

pip uninstall opencv-python opencv-python-headless
pip install opencv-python opencv-contrib-python

Acknowledgements

Code for reading binary COLMAP data is partly borrowed from COLMAP Utility Scripts by uzh-rpg
Thanks to Baptiste for providing the wooden block reconstruction data from [1]

References

[1] Erich, F., Bourreau, B., Neural Scanning: Rendering and determining geometry of household objects using Neural Radiance Fields. Link. 2022

Citation

If this work helps with your research, please cite:

@inproceedings{meyer2023cherrypicker,
  title={CherryPicker: Semantic skeletonization and topological reconstruction of cherry trees},
  author={Meyer, Lukas and Gilson, Andreas and Scholz, Oliver and Stamminger, Marc},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={6244--6253},
  year={2023}
}