Spatial Memory for AR Glasses: Hierarchical Visual Acquisition and Source-Linked Recall
| Source: arXiv AI
Tags: AR Glasses, Spatial Memory, Egocentric Vision, Wearables, Retrieval
A dual-camera AR-glasses prototype gives wearers searchable memory of past physical encounters by capturing color images selectively, taking just six RGB views from 648 observations in a 131-second replay.
Details
The system lets optical see-through AR glasses recall earlier encounters. A monochrome camera monitors constantly, and quality screening, appearance novelty and temporal support decide when to take a detailed RGB capture. Semantic observations keep links to their source images, so retrieval can answer indexed questions directly or go back to the image for details a scene description omitted. The design separates capture, interpretation and support for a specific question, with bounded reassessment of familiar views and retention guided by source dependencies. In a 131.35-second egocentric replay, the policy chose six RGB views from 648 observations, while feature extraction still ran on 97.7% of them, which the authors flag as a remaining cost.