About sobss

Compact building reconstruction based on Single-orientation Building Section Skeleton (BSS). This is an implementation of the paper:

Yijie Wu, Fan Xue*, Maosu Li, and Sou-Han Chen, 2024.
A novel Building Section Skeleton for compact 3D reconstruction from point clouds: A study of high-density urban scenes.
ISPRS Journal of Photogrammetry and Remote Sensing 209, 85-100. 

How to use

Dependencies

• open3d
• cgal
• nlopt
• eigen
• xtensor
• gurobi
• openmp
• rapidjson

Build from source

git clone https://github.com/eiiijiiiy/sobss.git
cd sobss
mkdir build
cd build
cmake ..
make

Run the gui

Video instruction (click the figure below to view)

Text instruction

• Set the path to the input point cloud, a directory to save the intermediates and final results, and parameters (optional).

cd ../
python main.py \
-i <path to the input point cloud *.ply [required]> \
-w <directory of the working folder (will be cleared before running) [required]> \
-v <voxel size in `skeletonizing` [optional]> \
-p <distance interval to group BSS atoms in `coarse segmentation` [optional]> \
-n <angular interval to group BSS atoms in `coarse segmentation` [optional]> \
-s <truncated distance in `merging` [optional]> \
-l <factor weighting the compactness and accuracy in `merging` [optional]> 

• Click the buttons skeletonize->coarse segment-> merge->evaluate. Results will be shown in the open3d window.

OS tested on

• Ubuntu 20.04 LTS
• MacOS 14.1.1 M1

Test data and results

Data and results can be downloaded at OneDrive:

• Inputs: 15 scenes tested in the paper can be found in 0-input.
• Results: Reconstructed results are in 2-union_tri(results).
• Evaluation: Inputs colorized in terms of the distances to the results are in 3-r2s(evaluation).

Contacts

• Wu, Y.: yijiewu@connect.hku.hk
• Xue, F.: xuef@hku.hk, frankxue.com