Deep Learning Based 6D Pose Estimation of Unknown Objects

6D pose estimation of unknown objects is an unsolved research problem in computer vision. This contribution proposes a novel method for estimating an object's pose using a monocular camera and an artificially created point cloud from the geometric object data. Point cloud generation and pose estimation are done with deep learning methods, combining several neural network-based approaches in one system. This paper describes the architecture and methods and compares and evaluates several ideas.