Volumetric Sensing and Streaming ยท MobiSys 2019

GLEAM: An Illumination Estimation Framework for Real-Time Photorealistic Augmented Reality on Mobile Devices

Authors
Siddhant Prakash, Alireza Bahremand, Linda D. Nguyen, Robert LiKamWa
Venue
MobiSys 2019 - The 17th International Conference on Mobile Systems, Applications, and Services

GLEAM captures the light already present in a scene and turns it into a usable illumination model, allowing mobile AR objects to share the visual character of their physical surroundings.

Introduction: Lighting That Belongs in the Scene

Virtual objects only feel integrated into augmented reality when their appearance responds to the physical environment. GLEAM estimates real-world illumination in real time, making reflective and light-sensitive materials such as glass, water, and metal more believable on mobile devices.

Physical elephant under real scene lighting
Real lighting.
Virtual elephant without illumination estimation
Without scene lighting.
Virtual elephant rendered with GLEAM illumination estimation
GLEAM illumination.

System Architecture: Radiance Samples to a Scene Cubemap

GLEAM observes reflective objects in the device camera to gather environmental radiance samples. It combines radiance sampling, optional transfer across devices, and cubemap composition to estimate lighting from all directions and apply it to virtual content.

GLEAM method overview showing radiance sampling network transfer and cubemap composition
Figure 1. GLEAM converts image-based radiance samples into an illumination cubemap for AR rendering.

Results: Balancing Fidelity and Adaptiveness

GLEAM can operate across a spectrum of update rates and cubemap resolutions, enabling responsive lighting for fast-changing scenes or higher-fidelity estimates when conditions allow. In user testing, the framework outperformed standard mobile AR lighting tools.

GLEAM runtime characterization result
Sampling performance.
GLEAM quality characterization result
Fidelity tradeoffs.
GLEAM update-rate characterization result
Adaptiveness tradeoffs.