Using PCL Gobal Descriptors in a DenseFusion Architecture

• Author(s): Pereira, Nuno ; Alexandre, Luís
• Date: 2019
• Persistent ID: http://hdl.handle.net/10400.6/8156
• Origin: uBibliorum
• Document Type: Conference object
• Language: English
• Contributor(s): uBibliorum

Description

In this paper, we present an alternative architecture to the state-of-the-art in 6D pose - DenseFusion. We changed the architecture of the method in the depth feature extraction phase. Instead of using the PointNet, as used in the original DenseFusion, we used global descriptors from the Point Cloud Library (PCL) to extract features. We made a comparison in terms of average accuracy between the Ensemble of Shape Functions (ESF), Viewpoint Feature Histogram (VHF) and the original PointNet.