This is the right place for you to get started!
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
README.md

README.md

Bayesian Object Tracking

The Bayesian Object Tracking organization on github is a collection of packages for 3D tracking of rigid objects (using depth images), as well as robot arm tracking (using depth images and joint encoders). For more details about the research which underlies these packages, please have a look at https://am.is.tuebingen.mpg.de/research_projects/probabilistic-object-tracking-using-a-depth-camera.

The core library for object tracking (dbot) is ROS independent. However, the integration with sensors (dbot_ros, dbrt) is based on the ROS eco-system.

Here, we give instructions on how to install the code and a getting started repository. This repository contains a complete example, including the necessary models and data. We recommend that you follow the instructions below to install and run the example. Once the example is working, you can adapt it to your needs.

Requirements

  • MS Kinect or Asus XTION depth sensor (unless you work with recorded data)
  • Ubuntu 14.04 (might work on other versions, but has not been tested)
  • c++11 Compiler (gcc-4.7 or later)
  • (optional) CUDA 6.5 or later with CUDA enabled graphic card

Dependencies

Object Tracking

The object tracking (dbot, dbot_ros) can be used without the robot tracking package (dbrt).

Workspace setup and compilation

cd $HOME
mkdir -p projects/tracking/src  
cd projects/tracking/src
git clone git@github.com:filtering-library/fl.git
git clone git@github.com:bayesian-object-tracking/dbot.git
git clone git@github.com:bayesian-object-tracking/dbot_ros_msgs.git
git clone git@github.com:bayesian-object-tracking/dbot_ros.git
cd ..
catkin_make -DCMAKE_BUILD_TYPE=Release -DDBOT_BUILD_GPU=On
source devel/setup.bash

If no CUDA enabled device is available, you can build without the GPU implementation via

catkin_make -DCMAKE_BUILD_TYPE=Release -DDBOT_BUILD_GPU=Off

Install and run the example

The getting started repository contains a ROS bagfile (a depth image sequence of an object being moved) and mesh models of some objects. Additionally it contains launch files, which allow you to run the code easily.

To install, follow these steps (note that cloning may take a while because the bagfile is large):

cd $HOME/projects/tracking/src
git clone https://git-amd.tuebingen.mpg.de/open-source/dbot_getting_started.git
cd ..
catkin_make -DCMAKE_BUILD_TYPE=Release -DDBOT_BUILD_GPU=On
source devel/setup.bash

Again, if no CUDA enabled device is available, you can build without the GPU implementation via

catkin_make -DCMAKE_BUILD_TYPE=Release -DDBOT_BUILD_GPU=Off

Now you can run the example:

roslaunch dbot_example launch_example_gpu.launch

If you did not install CUDA, you can run instead (note that the tracking performance is significantly better with the GPU version):

roslaunch dbot_example launch_example_cpu.launch

As soon as you launch the example, rviz should start, and an interactive marker should show up (in the form of an impact wrench). This marker is for initialization of the tracker, you can move it to align it with the point cloud. In this example, it should already be approximately aligned. Once you are done moving the marker, you can click on it and the tracker should start (note that in the recorded sequence the object starts moving at some point, make sure you initialize before that). You should see a green object model following the actual object visible in the white point cloud.

Additional documentation

For additional details about the object tracking, please checkout the dbot_ros package.

How to cite

inproceedings{wuthrich-iros-2013,
 title = {Probabilistic Object Tracking Using a Range Camera},
 author = {W{\"u}thrich, M. and Pastor, P. and Kalakrishnan, M. and Bohg, J. and Schaal, S.},
 booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
 pages = {3195-3202},
 publisher = {IEEE},
 month = nov,
 year = {2013},
 month_numeric = {11}
}

Robot Tracking

Workspace setup and compilation

The robot tracking setup builds on top of the object tracking, i.e. follow first the workspace setup of the object tracking above. Then continue with the instructions below:

cd $HOME/projects/tracking/src
git clone git@github.com:bayesian-object-tracking/dbrt.git
cd ..
catkin_make -DCMAKE_BUILD_TYPE=Release -DDBOT_BUILD_GPU=On

Again, if no CUDA enabled device is available, you can deactivate the GPU implementation via

catkin_make -DCMAKE_BUILD_TYPE=Release -DDBOT_BUILD_GPU=Off

Install and run the example

Add the following example project to the workspace (note that cloning may take a while due to the size of the data)

cd $HOME/projects/tracking/src
git clone https://git-amd.tuebingen.mpg.de/open-source/dbrt_getting_started.git
cd ..
catkin_make -DCMAKE_BUILD_TYPE=Release -DDBOT_BUILD_GPU=On
source devel/setup.bash

Now you can run the robot tracker along with the recorded sensory data:

roslaunch dbrt_example launch_example_gpu.launch

If CUDA is not being used, you can start the CPU based setup instead (note that the tracking performance is significantly better with the GPU version):

roslaunch dbrt_example launch_example_cpu.launch

This will start the data playback, the visualization and the robot tracker. You should see a point cloud in white, the robot model using only joint encoders in red, and the corrected robot model (fusing joint encoders and depth images) in blue. It should be visible that the blue robot model is significantly better aligned with the point cloud than the red one.

Additional documentation

For additional details about the robot tracking, please checkout the dbrt package.

How to cite

@article{GarciaCifuentes.RAL,
 title = {Probabilistic Articulated Real-Time Tracking for Robot Manipulation},
 author = {Garcia Cifuentes, Cristina and Issac, Jan and W{\"u}thrich, Manuel and Schaal, Stefan and Bohg, Jeannette},
 journal = {IEEE Robotics and Automation Letters (RA-L)},
 volume = {2},
 number = {2},
 pages = {577-584},
 month = apr,
 year = {2017},
 month_numeric = {4}
}