This code implements a semi-direct visual odometry pipeline that is described in the paper

• C. Forster, M. Pizzoli, D. Scaramuzza, "SVO: Fast Semi-Direct Monocular Visual Odometry," IEEE International Conference on Robotics and Automation (ICRA), 2014.

Video: http://youtu.be/2YnIMfw6bJY

SVO has been tested under ROS Groovy and Hydro and Ubuntu 12.04 and 13.04. This is research code, any fitness for a particular purpose is disclaimed.

The source code is released under a GPLv3 licence. A professional edition license for closed-source projects is also available. Please contact forster at ifi dot uzh dot ch for further information.

The API is documented here: http://uzh-rpg.github.io/rpg_svo/doc/

SVO can be used conveniently with ROS (www.ros.org). We use two workspaces, one for the plain CMake projects Sophus, Fast and optionally g2o and another workspace for the ROS-Catkin projects rpg_vikit and rpg_svo. Make sure to clone in the right folder.

Sophus by Hauke Strasdat implements Lie groups that we need to describe rigid body transformations. Checkout in your workspace for plain CMake projects.

cd workspace
git clone https://github.com/strasdat/Sophus.git
cd Sophus
git checkout a621ff
mkdir build
cd build
cmake ..
make

You don't need to install the library since cmake .. writes the package location to ~.cmake/packages/ where CMake can later find it.

The Fast detector by Edward Rosten is used to detect corners. To simplify installation we provide a CMake package that contains the fast detector from the libCVD library (http://www.edwardrosten.com/cvd/).

cd workspace
git clone https://github.com/uzh-rpg/fast.git
cd fast
mkdir build
cd build
cmake ..
make

Only required if you want to run bundle adjustment. It is not necessary for visual odometry. In fact, we don't run it on our MAVs. g2o requires the following system dependencies: cmake, libeigen3-dev, libsuitesparse-dev, libqt4-dev, qt4-qmake, libqglviewer-qt4-dev, install them with apt-get

I suggest an out-of-source build of g2o:

cd workspace
git clone https://github.com/RainerKuemmerle/g2o.git
cd g2o
mkdir build
cd build
cmake ..
make
sudo make install

If you don't want to make a system install, then you can replace the cmake command with cmake .. -DCMAKE_INSTALL_PREFIX:PATH=$HOME/installdir

vikit for contains camera models, some math and interpolation functions that SVO needs. vikit is a catkin project, therefore, download it into your catkin workspace source folder.

cd catkin_ws/src
git clone https://github.com/uzh-rpg/rpg_vikit.git

Sometimes cmake-modules is missing (required for including Eigen in ROS Indigo):

sudo apt-get install ros-hydro-cmake-modules

(replace hydro with your distribution).

Now we are ready to build SVO. Clone it into your catkin workspace

cd catkin_ws/src
git clone https://github.com/uzh-rpg/rpg_svo.git

If you installed g2o then set HAVE_G2O in svo/CMakeLists.txt to TRUE.

Finally, build:

catkin_make

Download this example dataset: rpg.ifi.uzh.ch/datasets/airground_rig_s3_2013-03-18_21-38-48.bag

Open a new console and start SVO with the prepared launchfile:

roslaunch svo_ros test_rig3.launch

Now you are ready to start the rosbag. Open a new console and change to the directory where you have downloaded the example dataset. Then type:

rosbag play airground_rig_s3_2013-03-18_21-38-48.bag

Now you should see the video with tracked features (green) and in RViz how the camera moves. If you want to see the number of tracked features, fps and tracking quality, run the GUI.

Type rosrun rqt_svo rqt_svo to run the SVO widget that displays the number of tracked features, the frame rate and provides some interface buttons.

If the widget is not found, try this:

rm ~/.config/ros.org/rqt_gui.ini
rosrun rqt_svo rqt_svo

Make sure to active the console window when pressing the keys.

• s Start/Restart
• q Quit
• r Reset

You need a calibrated camera to run SVO. We use the Matrix Vision Bluefox cameras in our lab, which have VGA resolution and global shutter. SVO supports three camera models:

1. The ATAN model - our preference - which is also used by PTAM. This model uses the FOV distortion model of "Deverneay and Faugeras, Straight lines have to be straight, 2001". You can calibrate your camera with this model by using the calibration tool in this package: https://github.com/ethz-asl/ethzasl_ptam We prefer this model because the projection and unprojection can be computed faster than with the other models. Further, the industral cameras that we use have neglectable tangential distortion.
2. The Pinhole model with three radial and two tangential distortion parameters. This model is standard in OpenCV and ROS. You can use the ROS camera calibration tool: http://wiki.ros.org/camera_calibration
3. The Ocam model by Davide Scaramuzza which can be used to model cameras with high field of view or even omnidirectional cameras. Use the OCamCalib toolbox to calibrate your camera: http://rpg.ifi.uzh.ch/software_datasets.html Although SVO can be used with this camera model, the algorithm does not work yet with omnidirectional cameras.

We save the calibration in YAML format (see examples in svo_ros/param) and include it in the launchfiles (see examples in svo_ros/launch).

A description of all parameters which can be set via the launchfile is provided in svo/include/config.h. The default parameters can be viewed in svo/src/config.cpp. Moreover, some additional parameters (mainly rostopic names etc.) are read from the ros parameter server in svo_ros/vo_node.cpp.

The current keyframe selection criterion is designed for downward looking cameras. This is one reason why SVO does not work well for forward motions (e.g., to be used on a car).

The current parameters are tuned for WVGA resolution. If you use a higher resolution camera, then the pyramid levels should be increased as well as the reprojection error thresholds.

You are very welcome to contribute to SVO by opening a pull request via Github. I try to follow the ROS C++ style guide http://wiki.ros.org/CppStyleGuide

If you use SVO in an academic context, please cite the following publication:

@inproceedings{Forster2014ICRA,
  author = {Forster, Christian and Pizzoli, Matia and Scaramuzza, Davide},
  title = {{SVO}: Fast Semi-Direct Monocular Visual Odometry},
  booktitle = {IEEE International Conference on Robotics and Automation (ICRA)},
  year = {2014}
}