Skip to content

Latest commit

 

History

History
162 lines (134 loc) · 6.88 KB

README.md

File metadata and controls

162 lines (134 loc) · 6.88 KB

kth-rgbd

Visual SLAM from RGB-D data
Copyright (C) 2011-2015 Virgile Högman
GNU GPL v3 License.

Mail URL: https://github.com/virgile-hogman/kth-rgbd/

Description

This library developed on Linux produces a 3D map from RGB-D data using Microsoft Kinect.
It includes a user-friendly program with a GUI to record RGB-D data and produce the 3D map in PCL format. Additional data about the map can be exploited by developers.
The SLAM approach is based on visual features such as SIFT and SURF.
To correct the inherent drift errors, loop closures are detected and the map is adjusted with a graph-oriented optimization (using g2o). It is strongly adviced to record loop closures or the optimization cannot be done.

License

Visual SLAM from RGB-D data
Copyright (C) 2011-2015 Virgile Högman

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program. If not, see http://www.gnu.org/licenses/.

Dependencies

There is no package distributed with kth-rgbd, so dependencies have to be installed correctly before building kth-rgbd.
OpenSIFT and g2o require manual installation (in local workspace, it is not necessary to install them at system level with root permissions).
The others dependencies should be installed with standard package managers.

g2o

g2o: A General Framework for Graph Optimization (OpenSLAM)
Build the repository version in trunk. Last tested with svn rev 54.
svn co https://svn.openslam.org/data/svn/g2o

NOTE! CSPARSE BUILD IN G2O
CSparse is an optional package in g2o which provides also a redistribution in EXTERNAL.
If CSparse is not found by g2o it builds its own version called libg2o_ext_csparse.so kth-rgbd uses this default behavior and looks for CSparse to know which headers and libs to use.
This default behavior can also be changed with BUILD_CSPARSE option (using ccmake) in g2o. A similar option G2O_BUILD_CSPARSE exists in kth-rgbd to force using EXTERNAL version built by g2o.
In any case kth-rgbd and g2o must be configured accordingly or there will be errors with the linker.

cd g2o/trunk
cd build
cmake ..
make

Note: the provided configuration kth-rgbd points to the build directory of g2o. It is not necessary to make install g2o at system level.

OpenSift

OpenSIFT Library - Rob Hess
(sub-dependencies: GTK2, OpenCV)
git clone https://github.com/robwhess/opensift

Build the (static) library using MAKEFILE.

cd opensift
make

Note: the provided configuration kth-rgbd point to the build directory of OpenSIFT. It is not necessary to make install OpenSIFT at system level.

Other dependencies

URLs are given for reference, install preferably with standard package managers.

  • OpenCV Last tested with 2.4.10
  • OpenNI Last tested with 1.5.7
    Requires OpenNI 1 (not OpenNI 2 SDK) preferably >=1.5
  • Point Cloud Libary Last tested with 1.7.2
    Requires PCL>=1.7 (sub-dependencies: cminpack, Flann)
  • Boost Library Last tested with 1.57
    Requires Boost>=1.46
  • Eigen3 Last tested with 3.0.2-3

Compilation

Get the source code from the main repository:
git clone https://github.com/virgile-hogman/kth-rgbd/

Edit kth-rgbd/CMakeLists.txt and set the correct paths for g2o and OpenSIFT.

set(g2o_SOURCE_DIR .../g2o/trunk)
set(OpenSIFT_DIR .../opensift)

The other packages should be found with standard cmake configuration files.

Build kth-rgbd using cmake:

cd kth-rgbd 	#your SVN checkout directory
mkdir build
cd build
cmake ..
make

No installation step is required.

Troubleshooting

  • if CSparse package is found by g2o then kth-rgbd may not link
    Configure g2o and kth-rgbd accordingly as described above. Not tested with SuiteSparse yet.
  • OpenNI headers like XnCppWrapper.h not found
    Be sure to use OpenNI1 (not 2). With OpenNI<1.5 headers are in /usr/include/openni
  • link problems with GTK and OpenSift
    Check instructions in CMakeLists.txt

Execution

No further configuration should be required if the scripts are used.
You can check the configuration file containing the parameters: kth-rgbd.cfg
The Kinect RGB-D camera should be properly setup before using the scripts.

Run the scripts from the script directory.
The user should launch the main script called record and move along with the camera slowly.
He should produce scenes with at least one cycle in order to create loop closures.
If a loop closure is detected, both initial and optimized map are produced.
If no loop closure is detected, only the initial map is produced and no optimization can be performed.

cd script
./record.sh

This will open a window with a GUI displaying the features but the registration does not start until the user decides it.
Commands can be launched from the window or console.

  • START: press any key. Frames are recorded in DIR_FRAMES.
  • STOP: press ENTER. This will create the map in DIR_PROD.
  • CANCEL: press ESC. Map is not created but the frames are kept.

DIR_FRAMES and DIR_PROD can be changed in scripts/var_env.sh

Short description of the scripts

  • record: main script, saves a sequence and builds map (with/without optimization depending on loop closures)
  • sequence: replays a sequence from given frames previously recorded
  • main: calls main program with all the custom arguments (check usage)
  • archive_results: saves the results into a folder
  • restore_results: restores the results previously saved
  • clean: removes current results

Known bugs

  • Main problem is residual drift on vertical axis (Y).
  • Enable ICP alignment.
  • Redistribution of g2o and OpenSIFT?

Contact

Virgile Högman
PhD Student at CVAP, KTH Royal Institute of Technology
virgile@kth.se