Atelier BMC (Background Models Challenge) à ACCV 2012, 5 Novembre 2012
25 Juin 2012
Catégorie : Conférence internationale
Nous organisons le 5 Novembre 2012 un atelier à ACCV portant sur l'évaluation d'algorithmes de soustraction de fond dans les vidéos. Vous trouvrez ci-dessous un descriptif de BMC (Background Models Challenge). Des premières vidéos et une première version du logiciel d'évaluation sont déjà disponibles sur le site bmc.univ-bpclermont.fr. Vous pouvez dès à présent paramétrer votre méthode grâce à ce premier matériel.
BMC - Background Models Challenge, ACCV 2012
Many real-time vision-based applications devoted to video surveillance search to distinguish moving objects from an image sequence given by a static camera. This background/foreground segmentation stage could be addressed by simple approaches, e.g. by computing the difference between two successive frames, or by building a time- averaged background image. However, such simple algorithms are very limited in outdoor environment (because of global variation of luminance, shadows of objects, etc.).
Since this is a popular topic, many background subtraction algorithms BSA) have been proposed since 90’s to tackle these problems. Although the evaluation of BSA is an important issue, relevant papers that handle with both benchmarks and annotated dataset are limited. Moreover, many authors that propose a novel approach use classic Gaussian mixture models as a gold-standard, but rarely compare their method with recent related. Background Models Challenge (BMC) is thus an opportunity for researchers from both universities and companies to evaluate the quality of their work.
- Submission of articles and results: July, 31st 2012
- Reviews and notification of acceptance: August, 20th 2012
- Camera ready articles and final results: October, 10th 2012
- BMC half-day: November, 5th 2012
- Antoine Vacavant, ISIT lab, CNRS/Univ. d'Auvergne, Clermont-Ferrand, France
- Laure Tougne, LIRIS lab, CNRS/Univ. Lumière Lyon2, Lyon, France
- Lionel Robinault, Foxstream company, Vaulx-en-Velin, France
- Thierry Chateau, LASMEA lab, CNRS/Univ. Blaise Pascal, Clermont-Ferrand, France
How to compete?
- Original and already published techniques can be submitted to BMC
- You can tune the parameters of your BSA thanks to the sequences of the "learning" phase (with ground truth), avalaible on this page
- More videos will be available soon, for the "evaluation" phase, still on this page
- Download the BMCW evaluation software from the website, and use it to compute quality measures
- Use the Excel file from BMC website to store your results, and write an article (6 pages max. with ACCV/LNCS model)
- Send your Excel file and your article to organizers
Articles accepted by the workshop will be contained in the proceedings of ACCV (LNCS format), and we expect to publish a special issue in Computer Vision & Image Understanding containing the three best papers, and one from an invited speaker.