GeneticMapFusion

A Python class for aligning and fusing two binary occupancy maps using a genetic algorithm. It includes additional methods for comparison using cross-correlation and ORB+RANSAC-based alignment.

Features

• Genetic algorithm-based optimisation for translation and rotation.
• Fitness function using Normalised Cross-Correlation (NCC) for robust evaluation of overlapping areas.
• Edge-weighted IoU available for additional metric analysis.
• Fused map generation after alignment.
• Alternative alignment methods:
  • Cross-correlation
  • ORB keypoints + RANSAC

Requirements

• Python 3.7+
• NumPy
• SciPy
• scikit-image
• OpenCV
• Matplotlib
• PIL (Pillow)
• IPython (for Jupyter notebook display)

Install dependencies:

pip install numpy scipy scikit-image opencv-python matplotlib pillow ipython

Usage

1. Initialisation

from genetic_fusion import GeneticMapFusion

ref_map = ...       # Binary 2D numpy array
target_map = ...    # Binary 2D numpy array

fusion = GeneticMapFusion(ref_map, target_map)

2. Run Evolutionary Alignment

best_params = fusion.evolve(
    tx_range=(-20, 20),
    ty_range=(-20, 20),
    theta_range=(-15, 15),
    fitness_threshold=0.80,
    min_generations=30,
    max_generations=1000
)

3. Fuse Maps with Optimal Parameters

fused_map, ref_canvas, target_canvas = fusion.fuse_maps_aligned(best_params)

4. Alternative Alignment Methods

• Cross-Correlation:

params_cc = fusion.align_with_cross_correlation()

• ORB + RANSAC:

params_ransac = fusion.align_with_ransac()

Output

• best_params: [tx, ty, theta] transformation to align target_map to ref_map.
• fused_map: Combined occupancy map with partial transparency (0.5) in overlapping areas.
• Optional visualisation using matplotlib.pyplot.imshow.

Notes

• Input maps must be binary numpy arrays.
• Performance depends on image quality, overlap, and initial parameter ranges.
• The evolutionary method is slower but more robust for maps with partial overlaps or low keypoint matches.

ROS 2 Node

We provide the ROS 2 node to test directly the algorithm reading occupancy grid maps from two different topics (robot 1 and robot 2). These topics are named /map1 and /map2, be sure to change the name if topics are named differently. Fused map is published to /fused_map

Node structure

map_fusion_ga/

├── map_fusion_ga/

│ ├── init.py

│ ├── fusion_node.py

│ └── genetic_fusion.py

├── launch/

│ └── fusion.launch.py

├── setup.py

├── package.xml

Build and Run

colcon build
source install/setup.bash
ros2 launch map_fusion_ga fusion.launch.py

License

CC by 4.0

Cite this!

@misc{https://doi.org/10.5281/zenodo.15696363,
  doi = {10.5281/ZENODO.15696363},
  url = {https://zenodo.org/doi/10.5281/zenodo.15696363},
  author = {Luna,  Cristina and Ruiz,  Enrique and López,  Paula and F. Barrero,  David},
  title = {Genetic Algorithm-based Map Fusion},
  publisher = {Zenodo},
  year = {2025},
  copyright = {Creative Commons Attribution 4.0 International}
}

Or: Luna, C., Ruiz, E., López, P., & F. Barrero, D. (2025). Genetic Algorithm-based Map Fusion (preliminary). Zenodo. https://doi.org/10.5281/zenodo.15696363