This project is based on the paper Towards an efficient 3D model estimation methodology for aerial and ground images which was published at Machine Vision and Applications Journal (2017). This code implements a Structure-from-Motion reconstruction pipeline described in the paper mentioned.
For more information, please acess the project page.
- Guilherme Augusto Potje - PhD. Student - UFMG - [email protected]
- Gabriel Dias Resende - Undergraduate Student (Former member of VeRlab) - UFMG - [email protected]
- Erickson Rangel do Nascimento - Advisor - UFMG - [email protected]
- Mario Fernando Montenegro Campos - Advisor - UFMG - [email protected]
Federal University of Minas Gerais (UFMG)
Computer Science Department
Belo Horizonte - Minas Gerais -Brazil
VeRLab: Vison and Robotic Laboratory
http://www.verlab.dcc.ufmg.br
- OpenCV
- Ceres Solver
- OpenMVG
- Exiv2
- Put the sfm_params.txt file in a desired location and change the image dataset path for the desired one.
- Create a directory inside the image path called 'result'
- Inside the 'result' dir, create the directories 'txt' 'visualize' 'models' 'undistorted'
- You can change the sfm_params.txt parameters according one's needs
- Call "./VerlabSFM [path_to_sfm_params] 1" for image registration and "./VerlabSFM [path_to_sfm_params] 2" for camera pose and sparse structrure estimation.
Obs: For fisheye distorted images (like ones taken with a GoPro) it is strongly recommended to calibrate the images and remove the distortion before using this pipeline on them.
If you are using it to academic purpose, please cite:
Potje, G., Resende, G., Campos, M., & Nascimento, E. R. (2017). Towards an efficient 3D model estimation methodology for aerial and ground images. Machine Vision and Applications, 28(8), 937-952.
@article{Potje2017,
title={Towards an efficient 3D model estimation methodology for aerial and ground images},
author={Potje, Guilherme and Resende, Gabriel and Campos, Mario and Nascimento, Erickson R},
journal={Machine Vision and Applications},
pages={1--16},
year={2017},
publisher={Springer}
doi = {10.1007/s00138-017-0875-x},
url = {https://link.springer.com/article/10.1007/s00138-017-0875-x}
}