Find file History
Pull request Compare This branch is 14 commits ahead, 1 commit behind MalcolmMielle:master.
Type Name Latest commit message Commit time
Failed to load latest commit information.
config bug fix for crashing and config file update to put model in home Jun 15, 2017
launch first commit Apr 22, 2016
src bug fix for crashing and config file update to put model in home Jun 15, 2017
srv removed backup files Jun 6, 2016
AUTHORS restructured the code-base Jun 6, 2016
CMakeLists.txt forked from malcolm and merged with gitlab Jun 6, 2016
LICENSE restructured the code-base Jun 6, 2016 restructured the code-base Jun 6, 2016
package.xml restructured the code-base Jun 6, 2016


This is a C++ implementation of OpenTLD that was originally published in MATLAB by Zdenek Kalal. OpenTLD is used for tracking objects in video streams. What makes this algorithm outstanding is that it does not make use of any training data. This implementation is based solely on open source libraries, meaning that you do not need any commercial products to compile or run it.

This version as PCL integration. Using a 3D camera, you can track a point in depth AND in position in the 2D image

Here is the list and description of all ROS parameters used in OpenTLD :

  • queue size : int, default 100. Set the queue size for the subscribers.

  • Tracking3D : bool, default true. Determine if OpenTLD needs only to do 2D tracking of the object (false) or if a full 3D Tracking using the Point Cloud to send informa- tion.

  • Graphical interface : bool, default true. Set the use of a graphical interface.

  • ShowTrajectory : bool, default false. If true, the trajectory of the object in the 2D image will be shown and recorded.

  • Trajectory length : int, default 50. If ShowTrajectory is set to true : length of the trajectory shown, in number of positions.

  • TimeBetweenFrames : double, default 0.1. Maximum time delay in between now and the frame used by openTLD to have a good latency.

  • LoadModel : bool, default false. Set to true, OpenTLD will try to load the model situated at the path provided by the next parameter, Path2Model.

  • Path2Model : string, default ” /model”. A path to a previously recorded model to be used by OpenTLD if LoadModel is set to True

  • SavingFile : string, default ” /model”. A path where to save the model currently used by OpenTLD, if requested by the user.

Here is a keyboard shortcuts :

  • l : toggle learning.

  • a : toggle alternating mode (if true, detector is switched off when tracker is available).

  • e : export Model.

  • i : import Model.

  • c : clear model, reinitialize the tracking.

  • r : ask the user for a new object in the current frame.

  • n : ask the user for a image from the current frame to add in the learning.