Spatio-temporal registration in augmented reality

Feng Zheng: Spatio-Temporal Registration in Augmented Reality (Under the direction of Gregory F. Welch) The overarching goal of Augmented Reality (AR) is to provide users with the illusion that virtual and real objects coexist indistinguishably in the same space. An effective persistent illusion requires accurate registration between the real and the virtual objects, registration that is spatially and temporally coherent. However, visible misregistration can be caused by many inherent error sources, such as errors in calibration, tracking, and modeling, and system delay. This dissertation focuses on new methods that could be considered part of “the last mile” of spatio-temporal registration in AR: closed-loop spatial registration and low-latency temporal registration: 1. For spatial registration, the primary insight is that calibration, tracking and modeling are means to an end—the ultimate goal is registration. In this spirit I present a novel pixel-wise closed-loop registration approach that can automatically minimize registration errors using a reference model comprised of the real scene model and the desired virtual augmentations. Registration errors are minimized in both global world space via camera pose refinement, and local screen space via pixel-wise adjustments. This approach is presented in the context of Video See-Through AR (VST-AR) and projector-based Spatial AR (SAR), where registration results are measurable using a commodity color camera. 2. For temporal registration, the primary insight is that the real-virtual relationships are evolving throughout the tracking, rendering, scanout, and display steps, and registration can be improved by leveraging fine-grained processing and display mechanisms. In this spirit I introduce a general end-to-end system pipeline with low latency, and propose an