![]() |
![]() |
---|---|
Ground-truth segmentation of a SynPEB scan | Corresponding PPE segmentation |
This repository contains the implementation of the plane extraction algorithm presented in the following paper:
Alexander Schaefer, Johan Vertens, Daniel Büscher, Wolfram Burgard:
A Maximum Likelihood Approach to Extract Finite Planes from 3-D Laser Scans.
Proceedings of the IEEE International Conference on Robotics and Automation, Montreal, Canada, May 2019.
[BibTeX]
This plane extraction algorithm takes an organized 3-D point cloud as input and outputs a set of planes that approximate the geometry underlying the point cloud.
To extract planes from an organized point cloud, use the function pcextrpln
.
For details about how to use pcextrpln
, please refer to the detailed documentation in the function header.
In order to accelerate the plane extraction process, it is recommended to compile the CUDA code in the gpufit
folder.
For instructions, see the corresponding readme file.
Make sure the resulting mex
files are on the MATLAB path.
To test the plane extractor, run the following code, which extracts planes from the test data we provide:
% Read the point cloud.
pc = pcread('../data/test_100.pcd')
% Extract planes.
res = pcextrpln(pc, 'e', 0.1, 'device', 'gpu')
% Visually compare ground truth and the computed segmentation.
plotseg(pc, res.plane)
While testing our plane extraction algorithm and comparing it to other methods, we found out that the well established SegComp dataset exhibits several deficiencies, which are described in our paper. In order to create a new strong baseline, we created the Synthetic Plane Extraction Benchmark: SynPEB. The SynPEB dataset can be downloaded at http:https://synpeb.cs.uni-freiburg.de/.
Side view of the ground-truth segmentation of a SegComp scan
To verify our experimental results, follow the steps below. They have been tested for MATLAB R2018a.
- Open MATLAB and navigate to the
matlab
folder. - Add all files in this folder to the MATLAB path.
- Run the script
gendata.m
to download the SegComp and SynPEB dataset. - Run the script
runexp.m
to perform the experiments. - To trigger the automatic evaluation of the data generated during the experiments, start the script
evalexp.m
. - Visualize the evaluation results by executing the script
ploteval.m
.