Depth Clustering

This is a fast and robust algorithm to segment point clouds taken with Velodyne sensor into objects. It works with all available Velodyne sensors, i.e. 16, 32 and 64 beam ones.

Check out a video that shows all objects which have a bounding box with the volume of less than 10 qubic meters:

How to build?

Prerequisites

Catkin.
OpenCV: sudo apt-get install libopencv-dev
QGLViewer: sudo apt-get install libqglviewer-dev
GLUT: sudo apt-get install freeglut3-dev
Qt (4 or 5 depending on system):
- Ubuntu 14.04: sudo apt-get install libqt4-dev
- Ubuntu 16.04: sudo apt-get install libqt5-dev
(optional) PCL - needed for saving clouds to disk
(optional) ROS - needed for subscribing to topics

Build script

This is a catkin package. So we assume that the code is in a catkin workspace and CMake knows about the existence of Catkin. Then you can build it from the project folder:

mkdir build
cd build
cmake ..
make -j4
(optional) ctest -VV

It can also be built with catkin_tools if the code is inside catkin workspace:

catkin build depth_clustering

P.S. in case you don't use catkin build you should. Install it by sudo pip install catkin_tools.

How to run?

See examples. There are ROS nodes as well as standalone binaries. Examples include showing axis oriented bounding boxes around found objects (these start with show_objects_ prefix) as well as a node to save all segments to disk. The examples should be easy to tweak for your needs.

Run on real world data

Go to folder with binaries:

cd <path_to_project>/build/devel/lib/depth_clustering

Frank Moosmann's "Velodyne SLAM" Dataset

Get the data:

mkdir data/; wget http://www.mrt.kit.edu/z/publ/download/velodyneslam/data/scenario1.zip -O data/moosmann.zip; unzip data/moosmann.zip -d data/; rm data/moosmann.zip

Run a binary to show detected objects:

./show_objects_moosmann --path data/scenario1/

Alternatively, you can run the data from Qt GUI (as in video):

./qt_gui_app

Once the GUI is shown, click on OpenFolder button and choose the folder where you have unpacked the png files, e.g. data/scenario1/. Navigate the viewer with arrows and controls seen on screen.

Other data

There are also examples on how to run the processing on KITTI data and on ROS input. Follow the --help output of each of the examples for more details.

Also you can load the data from the GUI. Make sure you are loading files with correct extension (*.txt and *.bin for KITTI, *.png for Moosmann's data).

Documentation

You should be able to get Doxygen documentation by running:

cd doc/
doxygen Doxyfile.conf

Related publications

Please cite related papers if you use this code:

@InProceedings{bogoslavskyi16iros,
title     = {Fast Range Image-Based Segmentation of Sparse 3D Laser Scans for Online Operation},
author    = {I. Bogoslavskyi and C. Stachniss},
booktitle = {Proc. of The International Conference on Intelligent Robots and Systems (IROS)},
year      = {2016},
url       = {http://www.ipb.uni-bonn.de/pdfs/bogoslavskyi16iros.pdf}
}

@Article{bogoslavskyi17pfg,
title   = {Efficient Online Segmentation for Sparse 3D Laser Scans},
author  = {I. Bogoslavskyi and C. Stachniss},
journal = {PFG -- Journal of Photogrammetry, Remote Sensing and Geoinformation Science},
year    = {2017},
pages   = {1--12},
url     = {https://link.springer.com/article/10.1007%2Fs41064-016-0003-y},
}

TmacTmac1992 / depth_clustering