Home People Research Publications Demos
News Jobs Prospective
About Internal

Saliency Based Discriminant Tracking

We consider tracking in the context of center-surround saliency mechanisms that are prevalent in biological vision. Discriminant tracking can be posed as a particular instance of this discriminant saliency: it provides a unified and principled framework for classifier design, target detection, automatic tracker initialization, and scale adaptation.

The video clips are in Quicktime format (H.264), and a most recent version of Quicktime is required. Quicktime is a free software and can be downloaded from here.

Comparison When Initial Target Position is Known















Error Rates (0 =perfect tracking; 1= total tracking failure)

Scale Adaptive Tracking





Average Tracking Error for IVT and DST

Automatic Initialization and Tracking






Comparison of Tracking Error with Automatic and Manual Initializations

  • Incremental learning for robust visual tracking, D. Ross, J. Lim, R. Lin, and M. Yang, In IJCV,
    77(1-3), pp 125-141, May 2008.
  • On-line selection of discriminative tracking features, R. Collins, Y. Liu, and M. Leordeanu, In IEEE PAMI, 27(10), pp 1631-1643, Oct 2005.
  • Ensemble tracking, S. Avidan,In IEEE PAMI, 29(2), pp 261-271, 2007.
  • Visual tracking with online multiple instance learning, B. Babenko, M.H. Yang, S. Belongie, In IEEE CVPR, 2009.


    © SVCL