zoe4751/ddpff

Plane Segmentation Using Depth Dependent Flood Fill

DDPFF

---

Accompanying code for the paper Plane Segmentation Using Depth-Dependent Flood Fill (pdf) by Arindam Roychoudhury, Marcell Missura and Maren Bennewitz accepted at IROS, 2021.

The source files containing the algorithm are DDPFF.h and DDPFF.cpp.

Build and execution intructions

1. Create a build folder (in source or out of source). Here we assume we create a build folder within the source directory: $mkdir build
2. Navigate to the build folder: $cd build
3. Execute cmake: $cmake ..
4. Execute the application: $./app/App

Application usage guide

A window appears when the application is executed.

Function	Navigation	Shortcut
Load point cloud frames	File → Load state	Ctrl+L
Navigate between frames	|<, <, >, >| buttons	Right and left arrow keys
Configuration	View → Config	C
Toggle point cloud	View → Point Cloud	P
Activate DDPFF algorithm	View → Depth dependent planar segments	Ctrl+D
Activate algorithm	View → Show Visualization	V
Activate merged planar segments	View → Show merged segments	Shift +U
Activate unmerged planar segments	View → Show unmerged segments	U
Show normals	View → Show plane normals	N
Activate camera view	View → Camera view	G
Activate ground truth planes	View → Show ground truth planes	Ctrl+G
Load Artificial Scenes	File → Load Scene	Ctrl+Shift+L
Choose scene	Scenes → Choose one of 6 scenes

The camera position in the artificial scenes can be modified by updating the scene parameters in the config widget. They can also be adjusted using the arrow keys and the Ctrl and Shift modifiers. Please select Scenes → Adjust Scene first to enable scene parameter updates using the arrow keys.

Parameter symbol-name mappings

Parameter symbol in paper	Parameter name in application config
κpoint	c_point
γpoint	pointThreshold_min
εpoint	pointThreshold_max
κflood	c_plane
γflood	planeThreshold_flood
εflood	planeThreshold_flood_max
κmerge	c_plane_merge
γmerge	planeThreshold_merge
εmerge	planeThreshold_merge_max
κangle	c_angle
γangle	angleThresholdFloodFill
εangle	angleThresholdFloodFill_max
κρ	c_range
γρ	normalSampleDistance_min
ερ	normalSampleDistance_max

The values of the parameters can be found in the paper.

Software requirements

1. C++ 17
2. Cmake 3.8 or greater
3. Qt 5.15
4. Eigen 3.3
5. OpenGL
6. QGLViewer
7. OpenCV 4
8. PCL 1.10

Using Docker

It is easiest to set up the development environment using docker. An image with the complete list of required softwares is available at dockerhub.

First, update the location of the source directory in the file docker/args.sh. Change the variable SOURCELOC to point to the source local directory of the repository.

Now, to create the development environment simply navigate to the directory "docker" and execute $./init_qtcreator_integrated.sh. This sets up a container with all the requisite softwares as well as launches the qtcreator IDE. This container shares the source directory and the data directory with the host system. Import the project into the IDE, build and execute it. To launch qtcreator subsequently use ./start_qtcreator_integrated.

Datasets

The application supports the following datasets:

Dataset	URL	Instructions
Kinect dataset[1]	http://www.ais.uni-bonn.de/download/segmentation/kinect.zip	Rename the archive to kinect.pcd.gt.zip and load.
SEGCOMP ABW dataset[2]	ftp://figment.csee.usf.edu/pub/segmentation-comparison/ABW-TEST-IMAGES.tar	Navigate to globals/include/globals/constants.h and change IMAGE_WIDTH and IMAGE_HEIGHT to 512. Recompile. Rename the archive to ABW-TEST-IMAGES.abw.tar and load.
TUM RGB-D Slam dataset[3]	https://vision.in.tum.de/data/datasets/rgbd-dataset/download	Each sequence which is a *.tgz file can be loaded individually.
Point clouds collected using Asus Xtion Pro Live	https://drive.google.com/file/d/1qqIUQoXyZsv3jmwHhzcJK8Msy8loSWrN/view?usp=sharing	Uncompress and load individual files.

Please use conf/configABW.conf when using the SEGCOMP ABW dataset. For the other datasets either conf/config5.conf (for higher accuracy) or conf/config10.conf (for higher performance) would work. Please rename each configuration files to conf/config.conf before using them.

References

[1]: Bastian Oehler, Joerg Stueckler, Jochen Welle, Dirk Schulz, and Sven Behnke. "Efficient Multi-Resolution Plane Segmentation of 3D Point Clouds" Proceedings of the 4th International Conference on Intelligent Robotics and Applications (ICIRA), Aachen, December 2011

[2]: A. Hoover, G. Jean-Baptiste, X. Jiang, P. J. Flynn, H. Bunke, D. B. Goldgof, K. Bowyer, D. W. Eggert, A. Fitzgibbon, and R. B. Fisher, "An experimental comparison of range image segmentation algorithms," IEEE transactions on pattern analysis and machine intelligence, vol. 18, no. 7, pp. 673–689, 1996.

[3]: J. Sturm, N. Engelhard, F. Endres, W. Burgard and D. Cremers, A Benchmark for the Evaluation of RGB-D SLAM Systems, Proc. of the International Conference on Intelligent Robot Systems (IROS), Oct. 2012.