Skip to content
A C++ implementation of OpenTLD
Pull request Compare This branch is 101 commits behind gnebehay:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.


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.

The easiest way to get started is to download the precompiled binaries that are available for Windows and Ubuntu 10.04. You can get a .deb package here: and a windows installer file here:

If you have a webcam attached to your PC, you can simply execute tld (on Ubuntu) or tld.exe (on Windows) in order to try it out. For other configuration options, please see below.


Keyboard shortcuts

b...remember current frame as background model / clear background
c...clear model and stop tracking
l...toggle learning
a...toggle alternating mode (if true, detector is switched off when tracker is available)
e...export model to file specified in configuration parameter "modelExportFile"
i...import model from file specified in configuration parameter "modelPath"
r...clear model, let user reinit tracking

command line options

tld [option arguments] [arguments]

option arguments
[-a <startFrameNumber>] video starts at the frameNumber startFrameNumber
[-b <x,y,w,h>] Initial bounding box
[-c] shows color images instead of greyscale
[-d <device>] select input device: device=(IMGS|CAM|VID)
IMGS: capture from images CAM: capture from connected camera VID: capture from a video [-e <path>] export model after run to path
[-f] shows foreground
[-i <path>] path to the images or to the video
[-h] shows help
[-j <number>] specifies the number of the last frames which are considered by the trajectory; 0 disables the trajectory
[-m <path>] if specified load a model from path. An initialBoundingBox must be specified or selectManually must be true.
[-p path] prints results into the file path
[-q] open QT-Config GUI
[-s] if set, user can select initial bounding box
[-t <theta>] threshold for determining positive results
[-z <lastFrameNumber>] video ends at the frameNumber lastFrameNumber. If lastFrameNumber is 0 or the optaion argument isn't specified means take all frames are taken.

[CONFIG_FILE] to the config file



  • OpenCV & OpenCV highgui >= 2.0
  • CMake >= 2.6
  • QtGui4 & QtCore4 >= 4.4 (optional)
  • NSIS (optional, Windows)


Use CMake to build the project. You can use "cmake-gui", if you need a graphical user interface.

  • CMake
    • Specify the source path (root path of the dictionary) and the binary path (where to build the program, out of source buid recommended)
    • Configure
    • Select compiler
    • Define the install path and other options
    • Generate
  • Windows: go to the binary directory and build the solutions you want (!!!You have to compile in RELEASE mode!!!):
    • tld: build the project
    • INSTALL: install the project
    • PACKAGE: create a NSIS-installer (NSIS required)
  • Linux: go with the terminal to the build directory
    • make: build the project
    • make install all: install the project
    • make package: generates a package (.tgz file if GENERATE_DEB_PACKAGE=false, .deb file if GENERATE_DEB_PACKAGE=true)


  • BUILD_WITH_QT: build project with a Qt-settings-dialog (Qt must be installed)
  • GENERATE_DEB_PACKAGE(Linux): build a Debian (.deb) packages
Something went wrong with that request. Please try again.