Transactions on Computational Science XIX pp 143-157
Pose Refinement of Transparent Rigid Objects with a Stereo Camera
We propose a new method for refining 6-DOF pose of rigid transparent objects. The algorithm is based on minimizing the distance between edges in a test image and a set of edges produced by the training model with a specific pose. The model is scanned with a monocular camera and a 3D sensor such as a Kinect device. The pose is estimated from a monocular image or a stereo pair. The method does not require a CAD model of the object. We demonstrate experimental results on a set of kitchen items essential for any home and office environment.
Keywordspose estimation localization transparent objects
Unable to display preview. Download preview PDF.