This repository contains a collection of Python scripts designed to automate various tasks using COLMAP, an open-source Structure-from-Motion (SfM) and Multi-View Stereo (MVS) software. These tools facilitate processes ranging from 3D reconstruction to camera pose estimation and 3D point reprojection onto 2D images.

Automates the full pipeline for 3D reconstruction using COLMAP, including database creation, feature extraction, feature matching, sparse reconstruction, and exporting the output.

Performs Perspective-n-Point (PnP) problem solving to estimate camera poses with an existing 3D model.

Provides utility functions for loading camera parameters, image data, and 3D point data from COLMAP's text output files.

Extracts camera extrinsic parameters from a COLMAP images.txt file for further analysis or processing.

Demonstrates how to reproject 3D points back onto 2D image planes using camera intrinsic and extrinsic parameters, including handling of quaternion to rotation matrix conversion and point distortion.

I have tested this code on Python 1.7.1.

To install the required Python packages, run the following command in your terminal:

conda install colmap
pip install -r requirements.txt

Make sure you have COLMAP installed on your system. For installation instructions, refer to the official COLMAP documentation.

Each script is designed to be run from the command line with specific arguments. Below are examples for each script:

python run_sfm.py --image_path <path_to_images> --colmap_data <path_to_colmap_data> --camera_model <camera_model>

python pnp.py --images <path_to_images> --database <path_to_colmap_database> --existing_reconstruction <path_to_existing_model> --output_path <output_path>

python extract_extrinsics.py --images_txt <path_to_images_txt> --output <output_pickle_file> --base <base_name>

python reproject.py --images <path_to_images> --colmap_output <path_to_colmap_output> --camera_id <camera_id> --output <output_path>

Contributions are welcome! If you have suggestions for improvements or bug fixes, please feel free to submit a pull request or open an issue.

This project is open source and available under the MIT License.