CTU Events

Stefan Gumhold - Object Pose Estimation and Tracking with 3D Object Coordinates

24 Oct 2014   12:45-13:30

This work addresses the problem of estimating and tracking the 6D Pose of specific objects from a single RGB-D image. We present a flexible approach that can deal with generic objects, both textured and texture-less. The key new concept is a learned, intermediate representation in form of a dense 3D object coordinate estimate paired with a dense object labelling. Random forests are exploited to predict 3D object coordinates and object probability from local image patches. We pair this discriminative estimate with a generative explanation of the RGB-D image in an energy functional that is robustly minimized with a RANSAC like approach. Tracking is based on the particle filter framework for which we designed efficient proposal distributions in order to efficiently sample the six dimensional parameter space of object poses. The proposed approach allows for real-time tracking and is very robust to changes in lighting and to occlusions.

Place

posluchárna KN:E-301 (ČVUT FEL, Karlovo náměstí 13, budova E)

Contact person

Ing. Jiří Bittner, Ph.D., bittner@fel.cvut.cz

More information

http://www.inf.tu-dresden.de/index.php?ln=en&node_id=3289