HL-MRF

HL-MRF

News

[2022-11-19] The paper has been published online! Check here to get it.
[2022-11-8] Our paper has been accpeted by ISPRS (International Society for Photogrammetry and Remote Sensing) !
[2022-5-27] Submit the manuscript to the journal.

Hierarchical Loop-based Multiview Registration Framework

HL-MRF can automatically perform the point-cloud-based registration for large-scale TLS scans without any artificial target. The main focus of our framework lies in the efficiency of processing large-scale TLS scans and being able to tolerate pairwise registration failure for robustness. The proposed hierarchical registration strategy and the loop-based coarse registration method deal with them respectively.

Two key features in HL-MRF are:

An internal scan-block and block-to-block registration strategy to hierarchically align point clouds of large-scale TLS scans.

A loop-based coarse registration method to remove grossly wrong pairwise registration results caused by ambiguous scanning content.

Demo

Pipline

ETH TLS Datasets

Robotic 3D Scan datasets

WHU TLS datasets

How to use

The code was tested on Windows (Visual Studio 2021) and Ubuntu 18.04

1. Pre-requisite

CMake version 2.8 or above
PCL version 1.11.0 (other versions may also work)
Boost version 1.65 or above

2. Compile

Clone this repository

git clone https://github.com/WuHao-WHU/HL-MRF.git
cd HL-MRF
mkdir build
cd build
cmake ..
make

3. run

On Ubuntu

cd ..
sh script/run.sh

On Windows, you can refer to run.sh to pass parameters in Visual Studio

4. parameter configuration

#./script/run.sh
# TLS input path
    path_dir=../scans/;#input directory
    out_dir=../result;#output directory
#parameters
	block_size=5;              # block size for hierarchical regisration (default: 5)
    downsample_size=0.1;       # downsample size of input TLS scans for coarse registration (default: 0.1 m)
	downsample_size_icp=0.1;   # distance of searching corresponding point in icp (default: the same as ${downsample_size})
	lum_iter=3;                # number of iteration for lum optimization (default: 3)
	t_MCS=10;                  # threshold of MCS (default: 10)
	number_of_threads=16;      # number of threads you use (default: 16)
	visualize_each_block=0;    # visualize each scan-block (default: 0)

5. Data preparation

You can test on the open-source TLS dataset: WHU-TLS, Robotic 3D Scan datasets, ETH Dataset.

You can also use your own data and edit the path_dir in run.sh.

The framework supports *.ply, *.pcd formats of point cloud data. You may need to transform other formats to the supported formats.

Paper

If you find out our code useful in your research, please cite

@article{WU202365,
          title = {A hierarchical multiview registration framework of TLS point clouds based on loop constraint},
          author = {Hao Wu and Li Yan and Hong Xie and Pengcheng Wei and Jicheng Dai},
          journal = {ISPRS Journal of Photogrammetry and Remote Sensing},
          volume = {195},
          pages = {65-76},
          year = {2023}
}

@article{Yan2022,
          archivePrefix = {arXiv},
          arxivId = {2205.07404},
          author = {Yan, Li and Wei, Pengcheng and Xie, Hong and Dai, Jicheng and Wu, Hao and Huang, Ming},
          eprint = {2205.07404},
          pages = {1--16},
          title = {A New Outlier Removal Strategy Based on Reliability of Correspondence Graph for Fast Point Cloud Registration},
          year = {2022}
}

@article{theiler2015globally,
          title={Globally consistent registration of terrestrial laser scans via graph optimization},
          author={Theiler, Pascal Willy and Wegner, Jan Dirk and Schindler, Konrad},
          journal={ISPRS journal of photogrammetry and remote sensing},
          volume={109},
          pages={126--138},
          year={2015},
          publisher={Elsevier}
}

Contributor

Hao Wu (吴豪)，Pengcheng Wei (韦朋成)

Contact

Email: [email protected]

Do not hesitate to contact the authors if you have any question or find any bugs.

yuancaimaiyi / hl-mrf Goto Github PK

hl-mrf's Introduction