Skip to content

This is a general and scalable framework for visual SLAM. It employs "Double Window Optimization" (DWO).

License

Notifications You must be signed in to change notification settings

Jinqiang/ScaViSLAM

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

42 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

***** ScaViSLAM *****

This is a general and scalable framework for visual SLAM. It employs 
"Double Window Optimization" (DWO) as described in our ICCV paper:

H. Strasdat, A.J. Davison, J.M.M. Montiel, and K. Konolige
"Double Window Optimisation for Constant Time Visual SLAM"
Proceedings of the IEEE International Conference on Computer Vision, 2011.

Check out the related videos: 
http://youtu.be/90Rw3qDuWrw
http://youtu.be/89CNVQ5azsc

At the moment it supports calibrated stereo rigs, and RGB-D cameras.
Monocular SLAM is NOT supported (yet).

*** NOTES AND WARNINGS / READ FIRST ***

This library is a research-level software which is in development. 
This is NOT a release version. For sure, features will be added, 
bugs will be fixed and the API will change in the future.

Even thought the software is in principal designed to run cross-platform,
it most likely will only work on a recent 64-bit linux (without the need
of adaptations...). It was tested on Ubuntu 11.10 - 64 bit.

(Note there might be problems on 32bit linux related to Eigen3.
 http://eigen.tuxfamily.org/dox/TopicUnalignedArrayAssert.html.
 I followed the guide describe on this website, but I have not tested it on 
 32-bit linux yet. Any bug reports/patches are welcome!)

WARNING: There is a bug concerning Eigen3.0 and gcc/g++-4.4. Make sure you install Eigen3.1 as explained below. 
(see: http://listengine.tuxfamily.org/lists.tuxfamily.org/eigen/2011/04/msg00013.html
&  http://comments.gmane.org/gmane.comp.lib.eigen/2046)
Detailed instructions are listed below.


*** INSTALLATION ***

First download sample image sequences. This will take a while, so do it first.
(These are already rectified stereo images from the New College dataset: 
http://www.robots.ox.ac.uk/NewCollegeData/)

>>
cd MY_DATA_DIRECTORY
mkdir newcollege
cd newcollege
wget -nH --cut-dirs=4 -r -l2 --no-parent -A "*.pnm" http://pub1.willowgarage.com/~konolige/data/NewCollege/FullData/StereoImages_1225720041.455302_to_1225720118.251935 
wget -nH --cut-dirs=4 -r -l2 --no-parent -A "*.pnm" http://pub1.willowgarage.com/~konolige/data/NewCollege/FullData/StereoImages_1225720118.301927_to_1225720193.248630
wget -nH --cut-dirs=4 -r -l2 --no-parent -A "*.pnm" http://pub1.willowgarage.com/~konolige/data/NewCollege/FullData/StereoImages_1225720193.298630_to_1225720268.945303
<<

If you want to download a larger portion of the data, see the index at

http://pub1.willowgarage.com/~konolige/data/NewCollege/FullData/

Before we start with the installation, we need to perform some preparations.
ScaViSLAM comes with GPU support. Indeed, GPU support is the default option.
(Each GPU method has a CPU equivalent. However those are not tested very much).
Thus, there is an optional dependency on CUDA 4.1 or higher.

Next, install dependencies:

>>
sudo apt-get install libeigen3-dev libsuitesparse-dev freeglut3-dev libglu-dev libglew-dev libboost-all-dev
<<

Important: Make sure libsuitesparse-dev is installed before you install g2o.

(If you use an older linux version, you might have to install eigen3 manually
from: http://eigen.tuxfamily.org)


In order to avoid version conflicts, I would very much recommend to install all  
libraries locally in your home directory and not system wide. So create 
a sub-directory in your home, e.g. 

>>
mkdir $HOME/svslocal 
<<

Now checkout/clone the ScaViSLAM repository.

>>
cd MY_SOFTWARE_DIRECTORY
git clone git://github.com/strasdat/ScaViSLAM.git
<<

Now, enter the project directory

>>
cd ScaViSLAM
<<

and then create and enter the directory for EXTERNAL libraries:

>>
mkdir EXTERNAL
cd EXTERNAL
<<

Check out and install g2o:

>>
git clone git://github.com/strasdat/g2o.git
cd g2o
mkdir svs_build
cd svs_build
cmake .. -DCMAKE_INSTALL_PREFIX:PATH=$HOME/svslocal
make -j4
make install
cd ../..
<<

Download and install opencv (version 2.4.2) inside EXTERNAL:

>>
wget http://sourceforge.net/projects/opencvlibrary/files/opencv-unix/2.4.2/OpenCV-2.4.2.tar.bz2
tar xvf OpenCV-2.4.2.tar.bz2
cd OpenCV-2.4.2
mkdir svs_build
cd svs_build
cmake .. -DCMAKE_INSTALL_PREFIX:PATH=$HOME/svslocal
make -j4
make install
cd ../..
<<

Clone and install Pangolin inside EXTERNAL

>>
git  clone git://github.com/strasdat/Pangolin.git
cd Pangolin
mkdir svs_build
cd svs_build
cmake .. -DBUILD_SHARED_LIBS=ON -DCMAKE_INSTALL_PREFIX:PATH=$HOME/svslocal
make -j4
make install
cd ../..
<<

Clone and install Sophus in folder EXTERNAL

>>
git clone git://github.com/strasdat/Sophus.git
cd Sophus
git checkout a621ff
mkdir svs_build
cd svs_build
cmake .. -DCMAKE_INSTALL_PREFIX:PATH=$HOME/svslocal
make -j4
make install
cd ../..
<<

Install VisionTools in folder EXTERNAL

>>
git clone git://github.com/strasdat/VisionTools.git
cd VisionTools
mkdir svs_build
cd svs_build
cmake .. -DCMAKE_PREFIX_PATH:PATH=$HOME/svslocal -DCMAKE_INSTALL_PREFIX:PATH=$HOME/svslocal
make -j4
make install
<<

Move to main project folder "ScaViSLAM" and install ScaViSLAM:

>>
cd ../../../
mkdir svs_build
cd svs_build
cmake .. -DCMAKE_PREFIX_PATH:PATH=$HOME/svslocal
make -j4
<<

In case you would like to compile ScaViSLAM WITHOUT CUDA support, 
turn CUDA_SUPPORT to OFF in the CMakeLists.txt file.

Make sure the "New College" data set is downloaded completely.
Then add a sym-link inside "ScaViSLAM/data":

>>
cd ../data
ln -s PATH_TO_MY_DATA_DIRECTORY/newcollege/(subdirectory) newcollege
<<

Where subdirectory contains the image sequence directory.  
If you want to run a longer sequence, consolidate the files into a
single directory.

Run ScaViSLAM on new college image sequence:
>>
cd ../svs_build
./stereo_slam ../data/newcollege.cfg
<<


*** HOW TO USE A RGB-D CAMERA? ***

First note, that the use of RGB-D cameras in ScaViSLAM is not very well tested and
very experimental! (ScaViSLAM is well tested in stereo-camera mode on the New College
seqeunce.)

At the moment, only RGB-D cameras following the PrimeSense reference design
(e.g. Ausus Xtion PRO LIVE) are supported.
First, you need to install PCL (which includes OpenNI):

>>
sudo add-apt-repository ppa:v-launchpad-jochen-sprickerhof-de/pcl
sudo apt-get update
sudo apt-get install libpcl-all
<<

Reinstall ScaViSLAM with PCL support:

>>
cd MY_SOFTWARE_DIRECTORY
cd /ScaViSLAM/
mkdir svs_build_with_pcl
cd svs_build_with_pcl
cmake  -DPCL_SUPPORT=ON ..
make -j8
>>

Try out ScaViSLAM with your RGB-D camera life:

>>
./stereo_slam ../data/rgbd-Live.cfg
<<

Dump RGB-D data to the hard disk (to ../data/out; create this directory or 
set an appropriate sym-link):

>>
./dump_rgbd ../data/rgbd_live.cfg 
<<

Finally, you can process the saved image sequence by adapting the configuration 
files "../data/rgbd_example.cfg". Basically, you have to point
"framepipe.path_str" to the saved images.


*** WHAT ABOUT LIVE STEREO CAMERAS? ***

This is more tricky since we need to make sure that the image stream is 
rectified. Some adaptation in the code might be necessary...


*** HOW TO CONTRIBUTE? ***

You are welcome to report bugs and request features at: 
https://github.com/strasdat/ScaViSLAM/issues.
To do so click on "New Issue", fill out the form and select the appropriate
label ("Bug" or "Feature Request").
Please feel free to also browse through the "ToDo" and "Future Features" 
issues, so you can find out which features are planned to be included 
soon ("ToDo") or eventually ("Future Features").

The best and easiest  way to contribute is to use the "github" way:
 - Register at github (for free).
 - Got to: "https://github.com/strasdat/ScaViSLAM"
 - Click on "Fork" to create a Fork of ScaViSLAM.
 - Commit and push changes into your Fork.
 - Click on "Pull Request" in order to ask me whether I would like to merge
   the changes into the ScaViSLAM master branch.

(If it is only a small change/bug fix, you could also simply send me
patch etc.)

In order to maximize the likelihood that I will accept your pull request,
please read carefully through the coding style guide:


*** CODING STYLE GUIDE ***

The coding style is largely influenced by the google c++ coding style guide:
http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml
(Reading the document is insightful in any case, even if you don't agree with
"google's" decision!)

I am especially strict about the following rules:

 - "When defining a function, parameter order is: inputs, then outputs."
 - "All parameters passed by reference must be labeled const."
   Thus, the following is NOT okay:
   'void method(const Vector2d & in, Vector2d & out)'
   Use this instead:
   'void method(const Vector2d & in, Vector2d * out)'
 - "Of the built-in C++ integer types, the only one used is int. If a program 
    needs a variable of a different size, use a precise-width integer type 
    from <stdint.h>, such as int16_t."
 - "Use 0 for integers, 0.0 for reals, NULL for pointers, and '\0' for chars."
 - "We do not use C++ exceptions."
 - "Use the specified order of declarations within a class: public: before 
    private:, methods before data members (variables), etc." 
 - "Use a struct only for passive objects that carry data; everything else 
    is a class."
 - "Prefer small and focused functions."
 - "Use only approved libraries and language extensions from C++11 (formerly 
    known as C++0x). Currently, none are approved."
 - "We do not allow default function parameters, except in a few uncommon 
    situations..."
 - "Use prefix form (++i) of the increment and decrement operators with 
    iterators and other template objects."
 - "We strongly recommend that you use const whenever it makes sense to do so."

At some points, I deviate from the C++ coding style guide:
 - 'Using namespace X;' is okay, but only within(!) the ScaViSLAM namespace.
   Never pollute the general namespace '::'. 
   (However, do not use "using namespace cv" or 'using namespace pangolin'.) 
 - I use 'boost::thread'.
 - In rare cases, it is okay to have public class members.
 - It is okay to use streams such as "cout", "cerr", "stringstream". (But 
   usually, I'd like to use "cerr" and friends only for debugging purposes...)

I also deviate slightly from google's formatting. Therefore the following rule
is the most important one:

"Use common sense and BE CONSISTENT.

If you are editing code, take a few minutes to look at the code around you and 
determine its style. If they use spaces around their if clauses, you should, 
too. If their comments have little boxes of stars around them, make your 
comments have little boxes of stars around them too." 

(All cites in "quotation marks" are from the google coding style guide.)

If you would like to use automatic indentation, I recommend you to use
the QtCreator (2.4.0 or newer) for code editing. You can import the ScaViSLAM
code style using "coding_style.xml":
 * Open "ScaViSLAM" project in QtCreator.
 * Click on "Projects". (Select "ScaViSLAM" tap in case several projects are 
   open.)
 * Click on "Editor Settings". 
 * Select "Editor Settings: Custom"
 * Click on "Coding Style Settings-->Import..." and select "coding_style.xml".

About

This is a general and scalable framework for visual SLAM. It employs "Double Window Optimization" (DWO).

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published