On performance analysis of optical flow algorithms

  • Authors:

  • Daniel Kondermann;Steffen Abraham;Gabriel Brostow;Wolfgang Förstner;Stefan Gehrig;Atsushi Imiya;Bernd Jähne;Felix Klose;Marcus Magnor;Helmut Mayer;Rudolf Mester;Tomas Pajdla;Ralf Reulke;Henning Zimmer

  • Affiliations:

  • Heidelberg Collaboratory for Image Processing, Interdisciplinary Center for Scientific Computing, University of Heidelberg, Heidelberg, Germany;Robert Bosch GmbH, Germany;University College London, United Kingdoms;Bonn University, Germany;Daimler AG, Germany;Chiba University, Japan;Heidelberg University, Germany;Technical University Braunschweig, Germany;Technical University Braunschweig, Germany;Bundeswehr University Munich, Germany;Linköping University (Sweden) and Goethe University, Frankfurt, Germany;Czech Technical University in Prague, Czech Republic;Humbold University Berlin, Germany;Saarland University, Germany

  • Venue:
  • Proceedings of the 15th international conference on Theoretical Foundations of Computer Vision: outdoor and large-scale real-world scene analysis
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

Literally thousands of articles on optical flow algorithms have been published in the past thirty years. Only a small subset of the suggested algorithms have been analyzed with respect to their performance. These evaluations were based on black-box tests, mainly yielding information on the average accuracy on test-sequences with ground truth. No theoretically sound justification exists on why this approach meaningfully and/or exhaustively describes the properties of optical flow algorithms. In practice, design choices are often made based on unmotivated criteria or by trial and error. This article is a position paper questioning current methods in performance analysis. Without empirical results, we discuss more rigorous and theoretically sound approaches which could enable scientists and engineers alike to make sufficiently motivated design choices for a given motion estimation task.