Relocalization using virtual keyframes for online environment map construction
Proceedings of the 16th ACM Symposium on Virtual Reality Software and Technology
| Hi-index | 0.00 |
For Augmented Reality (AR) technologies to experience wide-spread adoption, the barrier to entry must be reduced significantly. In particular, setup costs of new systems in new environments are prohibitive for high-fidelity augmentation. The goal of this thesis is to take first steps towards an overall reduction in these costs. The main AR challenges that I focus on in this thesis are realistic lighting of virtual geometry, registration of polygons with the scene, tracking a user in a mobile context, creation of 3D annotations, and acquisition of the scene lighting environment. Each of these challenges is a fundamental component of many AR systems that commonly requires laborious setup in new environments. The contributions presented in this thesis directly address these challenges, lowering the startup costs. My first contribution is to enable more realistic lighting in AR applications by dynamically measuring the lighting environment to accurately shade virtual objects, and to use the video image to facilitate application of virtual light sources to dynamic physical geometry. Realistic lighting also depends on accurate registration of polygons with physical objects. My second contribution reduces the impact of registration errors for these polygons by shifting polygon edges to match image intensity edges. For mobile AR systems, high quality wide-area tracking is necessary. My third contribution is the GroundCam, which combines a camera pointed at the ground, used as an optical mouse, with a GPS unit for untethered, high quality position estimation. Good position estimates are needed to accurately create 3D annotations in the outdoor scene, which is what my fourth contribution focuses on. The aerial annotator system uses aerial photographs as a second viewpoint to enable easy annotation placement in outdoor scenes. Finally, my fifth contribution is Envisor, an application that allows users to quickly acquire environment maps of these scenes with just a head-worn or hand-held camera. To facilitate the work in this thesis, I developed two frameworks for conducting AR research. ARWin is an indoor desktop system that acts as a 3D application window manager, and is rapidly-deployable in new environments. Outdoor, mobile research is conducted within ARagorn, a wearable computing platform built with inexpensive, commonly-available hardware.