Agrégation d'estimations semi-locales pour le flot optique

Abstract : Most existing methods for optical flow estimation rely on a global variational framework suffering from several limitations like loss of small structures with large displacements or over-smoothing of motion discontinuities. We present a novel optical flow estimation method based on a two-stage aggregation framework and designed to handle these issues. First, semi-local candidates are estimated with a combination of patch correspondences and affine motion estimations. Then, the candidates are combined in an aggregation stage creating a dense global flow field. We propose two distinct aggregation methods : the first one operates in a discrete framework and selects one candidate at each pixel with a graph-cut based optimization; the second one combines sparse continuous optimization and confidence measure and tolerates deviations from the candidates. Both approaches are free of coarse-to-fine schemes. Improvements on the aforementioned complex motion configurations are demonstrated on challenging examples.
Document type :
Conference papers
Complete list of metadatas

https://hal.inria.fr/hal-00931568
Contributor : Charles Kervrann <>
Submitted on : Wednesday, January 15, 2014 - 2:41:59 PM
Last modification on : Wednesday, April 11, 2018 - 1:53:35 AM

Identifiers

  • HAL Id : hal-00931568, version 1

Collections

Citation

Denis Fortun, Patrick Bouthemy, Charles Kervrann. Agrégation d'estimations semi-locales pour le flot optique. Gretsi - XXIVe Colloque Gretsi - Traitement du Signal et des Images - 2013, Sep 2013, Brest, France. pp.1-4. ⟨hal-00931568⟩

Share

Metrics

Record views

151