Home

Awesome

Hits

CoBigICP: Correntropy based Bidirectional Generalized ICP

CoBigICP is a fast and accurate point cloud registration method, currently implemented in MATLAB.

About

<img src="https://github.com/Pamphlett/CoBigICP/blob/master/assets/cmp.png" width="1000" div align=center>

Reconstruction results of ETH Hauptgebaude generated by CoBigICP and GICP. Both estimated aligned point clouds are colored, with the ground truth being white.

CoBigICP can solve 3D rigid transformation between two point clouds. It adopts correntropy as the noise model and prunes putative nearest neighbor by a bidirectional search step. Also a symmetric objective function is used to gather surface geometric information. For more information, please refer to our paper. (pdf)

Registration Example

Git clone or download the repo and add to your MATLAB workspace.

Run CoBigICP_fun.m to align two scans from ETH Hauptgebaude.

Paper

If you find the paper helpful or use the code in your project, please cite:

@inproceedings{cobigicp2020yin,
    title={CoBigICP: Robust and Precise Point Set Registration using Correntropy Metrics and Bidirectional Correspondence},
    author={Yin, Pengyu and Wang, Di and Du, Shaoyi and Ying, Shihui and Gao, Yue and Zheng, Nanning},
    booktitle={IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
    year={2020},
    organization={IEEE}
}

Acknowledgements

Many thanks to Dr.Ga for the fruitful discussions.