Séminaire de M. Vincent SPRUYT, IPI, Université de Ghent, Belgique - Date : vendredi 14 février 2014, 15h00 - Lieu : "seminar room"; 9ème étage, B1, MICA Institute, Hanoi University of Science and Technology

Intervenant :
M. Vincent SPRUYT, Groupe IPI (Image Processing and Interpretation), Université de Ghent, Belgique

Date : vendredi 14 février 2014, 15h00
Lieu : salle "seminar room", 9ème étage, bâtiment B1, MICA Institute, Hanoi University of Science and Technology
Langue : le séminaire sera présenté en anglais

Résume/abstract:
Real-time detection of human hands in undefined posture and in unconstrained environments is a challenging problem. In the first part of my presentation, I will present a parts-based object detector that is able to solve this problem. The detected hands are treated as hand hypotheses that are then verified by a particle filter tracking algorithm. I present several ways to improve traditional particle filtering methods by enhancing their proposal distribution.
Finally, I present a simple segmentation method to obtain a rough estimate of the hand contour. The detector, tracker and segmenter are combined in a closed-loop configuration such that they are able to adapt to changing illumination or backgrounds, and to learn from past mistakes.