Locally Orderless Tracking

Shaul Oron, Aharon Bar-Hillel, Dan Levi, Shai Avidan


Locally Orderless Tracking (LOT) is a visual tracking algorithm that automatically estimates the amount of local (dis)order in the object. This lets the tracker specialize in both rigid and deformable objects on-line and with no prior assumptions. We provide a probabilistic model of the object variations over time. The model is implemented using the Earth Moverís Distance (EMD) with two parameters that control the cost of moving pixels and changing their color. We adjust these costs on-line during tracking to account for the amount of local (dis)order in the object. We show LOTís tracking capabilities on challenging video sequences, both commonly used and new, demonstrating performance comparable to state-of-the-art methods.


  • Locally Orderless Tracking, PDF BIB
    Shaul Oron, Aharon Bar-Hillel, Dan Levi, Shai Avidan.
    International Journal on Computer Vision 2014.
  • Locally Orderless Tracking, PDF BIB
    Shaul Oron, Aharon Bar-Hillel, Dan Levi, Shai Avidan.
    Computer Vision and Pattern Recognition 2012.

Tracking Results

The first video demonstrates on-line parameter estimation showing the appearance noise variance σA and localization noise variance σL over time. The additional videos show tracking results of Locally Orderless Tracking (LOT) marked in Green compared with 4 state-of-the-art methods: Incremental Visual Tracker (IVT) marked in Yellow Online AdaBoost (OAB) Cyan Multiple Instance Learning (MIL) Red Visual Tracking Decomposition (VTD) marked in Magenta






We provide the Matlab source code used to obtain the results presented in the paper.


  • LOT Full Dataset
    The dataset provided here includes sequences taken from many publicly available databases and websites (see below). In addition we provide new and more accurate per-frame annotations for these sequences.
    To reproduce the results reported in the paper follow the instructions in the README file.

    Index of re-distributed sequences:
    1) david, face and sylv were taken from Boris Babenkos' MILTrack web-page
    2) shop, human and girl were taken from Haibin Lings' web-page
    3) dog was taken from David Ross' IVT web-page
    4) train was taken from PETS 2006
    5) lemming was taken from the PROST dataset
    6) ucsdpeds was taken from the UCSD-SVCL Crowd database
    7) skating was taken from Junseok Kwon & Kyoung Mu Lee VTD webpage

  • Presentations

    • Presentation given at the EE department seminar PPT
      Embedded videos might not play in Powerpoint versions prior to 2010


    Tracking , EMD , Joint spatial appearance , LOT , Locally Orderless