Skip to content

karlpauwels/simtrack

Repository files navigation

SimTrack

Build Status

SimTrack, a simulation-based framework for tracking, is a ROS-package for detecting and tracking the pose of multiple (textured) rigid objects in real-time. SimTrack is released under the BSD-license. SimTrack uses SiftGPU for feature extraction. Note that SiftGPU has a different license. Please cite the following paper if you use SimTrack in your research:

Pauwels, Karl and Kragic, Danica (2015) SimTrack: A Simulation-based Framework for Scalable Real-time Object Pose Detection and Tracking. IEEE/RSJ International Conference on Intelligent Robots and Systems, Hamburg, Germany, 2015.

For more details see the following paper:

Pauwels, Karl; Rubio, Leonardo; Ros, Eduardo (2015) Real-time Pose Detection and Tracking of Hundreds of Objects. IEEE Transactions on Circuits and Systems for Video Technology, in press.

Please have a look at these example videos of SimTrack in action:

System Requirements

  • Ubuntu 16.04 (use the indigo branch for Ubuntu 12.04 or 14.04)
  • Ros Kinetic (use the indigo branch for ROS Hydro or Indigo)
  • Monocular camera or RGB-D sensor (Asus Xtion, Primesense, Kinect v1 or Kinect v2)
  • Installed and working NVIDIA CUDA driver and toolkit (version 6.5 or above)
  • CUDA-capable Graphics Processing Unit (Fermi class or above)
    • Two GPUs are recommended
    • Xtion-style sensors (640x480) require at least 1.3 GB of free GPU memory (300 MB tracking, 1 GB detection) for the three example objects. Use nvidia-smi to check memory usage.
    • Kinect v2 sensors (1920x1080) require at least 2.6 GB of free GPU memory (400 MB tracking at QHD, 2.2 GB detection at full HD). The resolution can be lowered for detection but this is not ideal.

Installation

Source your ROS environment:

source /opt/ros/kinetic/setup.sh

Create your workspace:

mkdir -p ~/my-ws/src

Fetch the code:

cd ~/my-ws/src
git clone https://github.com/karlpauwels/simtrack.git

Install the dependencies:

cd ~/my-ws
sudo rosdep init # only if never run before
rosdep update
rosdep install --from-paths src --ignore-src -y -r

Build:

cd ~/my-ws
catkin_make -DCMAKE_BUILD_TYPE="Release"

Test

Check nvidia-settings and verify that Sync to VBlank is disabled under OpenGL Settings.

Initialize the environment:

cd ~/my-ws
source devel/setup.bash

Build the SIFT-model of the THREE provided demo objects. Note that an absolute path is required:

rosrun interface cmd_line_generate_sift_model `pwd`/src/simtrack/data/object_models/ros_fuerte/ros_fuerte.obj
rosrun interface cmd_line_generate_sift_model `pwd`/src/simtrack/data/object_models/ros_groovy/ros_groovy.obj
rosrun interface cmd_line_generate_sift_model `pwd`/src/simtrack/data/object_models/ros_hydro/ros_hydro.obj

This process should display rotated versions of the objects with SIFT keypoints highlighted. Three new files will be created in the respective model folders: ros_fuerte_SIFT.h5 (~4.2MB), ros_groovy_SIFT.h5 (~7.0MB) and ros_hydro_SIFT.h5 (~2.9MB).

Print models_to_print.pdf on three A4-pages. Check that the dimensions are correct using the printed ruler.

Adjust the GPU configuration in parameters.yaml to your system:

simtrack/tracker/device_id: 0
simtrack/detector/device_id: 1

For example, set both device_id's to 0 for a single-GPU system.

The default configuration uses openni_2-launch which supports Asus Xtion and Primesense devices. To use a Kinect, point main.launch to the camera_kinect.launch file.

Run SimTrack:

roslaunch simtrack_nodes main.launch

or if you're using a Kinect v2, and have the driver installed:

roslaunch simtrack_nodes main_kinect2.launch

The tracker output is available on the /simtrack/image topic. It can be adjusted through dynamic_reconfigure to display either the tracker state or the optical flow.

If all goes well, the printed pages should be detected and tracked! See the tutorial for help on modeling and tracking your own objects.

Tutorial

www.karlpauwels.com/simtrack

About

A simulation-based framework for tracking

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published