tum-vision / dvo

Dense Visual Odometry

Dense Visual Odometry (dvo)

These packages provide an implementation of the rigid body motion estimation of an RGB-D camera from consecutive images.

• dvo_core

  Core implementation of the motion estimation algorithm.

• dvo_ros

  Integration of dvo_core with ROS.

• dvo_benchmark

  Integration of dvo_core with TUM RGB-D benchmark, see http://vision.in.tum.de/data/datasets/rgbd-dataset.

• sophus

  ROS package wrapper for Hauke Strasdat's Sophus library, see https://github.com/strasdat/Sophus.

Installation

Checkout the branch for your ROS version into a folder in your ROS_PACKAGE_PATH and build the packages with rosmake.

• ROS Fuerte:

  git clone -b fuerte git://github.com/tum-vision/dvo.git
  rosmake dvo_core dvo_ros dvo_benchmark

• ROS Electric:

  You need to install perception_pcl_unstable with PCL version 1.5+.

  git clone -b electric git://github.com/tum-vision/dvo.git
  rosmake dvo_core dvo_ros dvo_benchmark

Usage

Estimating the camera trajectory from an RGB-D image stream:

• Start the OpenNI camera driver: roslaunch openni_launch openni.launch
• Start the dvo camera_tracker node: rosrun dvo_ros camera_tracker
• Start dynamic_reconfigure GUI
  • In /camera/driver enable depth_registration on
  • In /camera_tracker enable reconstruction, use_weighting, run_dense_tracking, and use_dense_tracking_estimate

If everything works, the stdout of the camera_tracker node should show [ WARN] [1355131430.132265592]: RGB image size has changed, resetting tracker! and the camera pose is published on the /rgbd/pose topic. You can restart the camera motion estimation by disabling and enabling the run_dense_tracking option.

For visualization:

• Start RVIZ
• Set the Target Frame to /world
• Add an Interactive Marker display and set its Update Topic to /dvo_vis/update
• Add a PointCloud2 display and set its Topic to /dvo_vis/cloud

The red camera shows the current camera position. The blue camera displays the initial camera position.

Publications

The following publications describe the approach:

• Robust Odometry Estimation for RGB-D Cameras (C. Kerl, J. Sturm, D. Cremers), In Proc. of the IEEE Int. Conf. on Robotics and Automation (ICRA), 2013
• Real-Time Visual Odometry from Dense RGB-D Images (F. Steinbruecker, J. Sturm, D. Cremers), In Workshop on Live Dense Reconstruction with Moving Cameras at the Intl. Conf. on Computer Vision (ICCV), 2011.

License

The packages dvo_core, dvo_ros, and dvo_benchmark are licensed under the GNU General Public License Version 3 (GPLv3), see http://www.gnu.org/licenses/gpl.html.

The package sophus is licensed under the MIT License, see http://opensource.org/licenses/MIT.