Low Latency Mobile Augmented Reality with Flexible Tracking

Abstract

Jaguar is a mobile Augmented Reality (AR) framework that leverages GPU acceleration on edge cloud to push the limit of end-to-end latency for AR systems and enable accurate and large-scale object recognition based on image retrieval. It integrates the emerging AR development tools (e.g., ARCore and ARKit) into its client design for achieving flexible, robust and context-aware object tracking. Our prototype implementation of Jaguar reduces the end-to-end AR latency to ~33 ms and achieves accurate six degrees of freedom (6DoF) tracking. In this demo, we will show that our Jaguar client recognizes movie posters within the camera view by offloading computation intensive tasks to edge cloud and augments these posters with their movie trailers in 3D upon receiving the recognition results.