MODEL-BASED 6D OBJECT POSE ESTIMATION

Abstract

6D object pose estimation is an important problem in computer vision with applications in robotic manipulation and augmented reality. Substantial progress has been achieved in the past decade due to deep learning techniques. However, it is still an open problem. In this thesis, we focus on three of the remaining challenges, i.e., pose ambiguity raised by symmetric objects, pose estimation of unseen objects, and high cost of labeling 6D poses. Each of the challenges is addressed with an efficient algorithm. Extensive experiments demonstrate that our algorithms are effective and outperform existing methods.