Cloud-based mobile augmented reality, pitfalls and strategies in a realtime deployment
Justin Manweiler · 2014
With the rise of Google Glass, etc., the era is now ripe for deeply context-aware Mobile Augmented Reality. With well-positioned cameras, 4G(+) connectivity, and public clouds, continuous (strong) computer vision is a possibility. Many research prototype mobile augmented reality systems assume and leverage a powerful backend system to alleviate resource-constrained mobile devices, often prescribed to be a ``cloud" service. This architecture, while promising and useful, introduces new complications to a practical deployment. For augmented reality, latency is a key limiting factor to the user's perception of quality. Latency, however, is nontrivial even with the massive, worldwide public cloud deployments available today. A second limiting factor is cost. Computational resources while ``cheap" in public clouds are not free. In this talk, I will discuss an example Mobile Augmented Reality system that deploys a realtime, latency-sensitive (AR) application with cloud backend processing in the critical path. In this system, we seek to dampen these challenging effects through a hybrid mobile-plus-cloud architecture that leverages both sensing and cloud-based strategies for mitigation.