Abstract
In this paper, we propose Point Decomposition Network (PointDCCNet) for 3D object categorization using point cloud decomposition. In the recent technologies for 3D data capture, point clouds have a surge in demand due to their simpler representation and computations. The point cloud analysis requires robust methods for feature extraction to tackle the permutation invariance and unorderdness in point sets and finds application in categorization, refinement, and super-resolution of 3D data. We propose a novel PointDCCNet towards the decomposition of point clouds into primitive geometric shapes, namely plane, sphere, cone and cylinder; and use it as a clue towards modelling a classifier for 3D object categorization. The decomposition of point clouds provides a geometrical signature of the 3D object towards categorization. We show the decomposition of 3D data into primitive shapes which assists the model in the categorization of 3D objects. We demonstrate the results using benchmark datasets and compare them with state-of-the-art techniques.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have