Energy Minimization Methods in Computer Vision and Pattern Recognition

Volume 3757 of the series Lecture Notes in Computer Science pp 203-219

Linear Programming Matching and Appearance-Adaptive Object Tracking

  • Hao JiangAffiliated withSchool of Computing Science, Simon Fraser University
  • , Mark S. DrewAffiliated withSchool of Computing Science, Simon Fraser University
  • , Ze-Nian LiAffiliated withSchool of Computing Science, Simon Fraser University

* Final gross prices may vary according to local VAT.

Get Access


In this paper, we present a novel successive relaxation linear programming scheme for solving the important class of consistent labeling problems for which an L 1 metric is involved. The unique feature of the proposed scheme is that we use a much smaller set of basis labels to represent the label space. In a coarse to fine manner, the approximation improves during iteration. The proposed scheme behaves very differently from other methods in which the label space is kept constant in the solution process, and is well suited for very large label set matching problems. Based on the proposed matching scheme, we develop a robust multi-template tracking method. We also increase the efficiency of the template searching by a Markov model. The proposed tracking method uses a small number of graph templates and is able to deal with cases in which objects change appearance drastically due to change of aspect or object deformation.