On 3D Pose Estimation for XR. Classic Approaches vs Neural Networks

Abstract

The primary purpose of this paper is to investigate different approaches for 3D object pose estimation, which uses neural networks, and for model-based tracking - an innovative solution that builds upon a combination of known matching and pose estimation algorithms and to propose the one which will be more suitable for our problem. Object tracking is one of the critical problems for many applications on AR/MR devices that use object pose estimation to create an immersive experience by combining the physical world with virtually generated objects. The main limitation of our application is that it must work in real-time and be efficient enough to run on devices with weak computing power (e.g., RealWear HMT-1).