Computer Vision for Remote AR
summary
Describes a production pipeline that turns phone video into a real-time textured 3D mesh for remote AR, using motion stereo (and LiDAR when available), plus a classifier over OCR output to extract model and serial numbers. Claims prior research boosted 2D→3D retrieval accuracy ~30% and reports a working Streem-integrated platform (Selerio acquired) tested by real customers. Tackles hard problems: temporal-consistent depth from 2D, texture updating without artifacts, and 6-DoF pose refinement using ML keypoint initialization followed by region-based optimization.
AI-written summary · about
chapters (10)
- 0:00Introduction and welcome
- 1:08Guest introduction and background
- 9:25Talk overview and key points
- 10:29Remote AR and computer vision challenges
- 14:45Real-time 3D meshing and texturing
- 19:23Extracting device metadata
- 24:436DoF pose estimation for tutorials
- 32:39Spatial AI and future vision
- 38:08ARVR workshop insights
- 40:13Closing and acknowledgments
related talks
43mBeyond The Printed Page
39mBig AR Talk: Three Innovators Discuss Unlocking AR's Potential
48mRendering techniques for augmented reality and a look ahead at AR foundation
15mVideo-Annotated Augmented Reality Assembly Tutorials
17mEnabling AR Everywhere: An Overview of DAQRI Products
42mHow to Augment Your Reality
26mFrom Panoramic Vision to Lightweight 3D Worlds for Semantic Visual Tracking
18m4 Keys to Augmented Reality’s Future