Work

Real-Time AR Glasses for Egocentric Vision: A Structured Review and Taxonomy

Year: 2026
Type: preprint
Abstract: Wearable augmented reality (AR) glasses combine head-mounted displays, forward-facing sensors, and computer vision to present context-aware information from the wearer’s first-person viewpoint. Their usefulness depends not only on recognition accuracy, but also on end-to-end latency, registration stability, power consumption, usability, and responsible handling of continuously captured data. This paper presents a structured review of research relevant to real-time AR glasses for egocentric vision. It examines wearable hardware and sensing, on-device and edge-assisted computation, visual perception, temporal understanding, interaction techniques, evaluation practices, and privacy considerations. Representative systems include the DARLENE wearable AR platform, an AI-assisted first-aid guidance prototype, and recent egocentric vision-language assistants. The review also considers foundational egocentric datasets and identifies limitations that complicate direct comparison between published systems. Rather than treating isolated model speed as a measure of real-time performance, the paper argues for end-to-end evaluation that includes sensing, inference, communication, rendering, and human response. The resulting taxonomy and design considerations are intended to support the development of responsive, resource-aware, and privacy-conscious wearable AR applications.
Language: English
Open Access status: green
OpenAlex ID: W7234627092
Cites:
Cited by:
Related to: 10