P‐4.27: Towards Low‐Power Image Feature Extraction and Matching on AR Glasses
SID Symposium Digest of Technical Papers · 2026-08 · 57 (S1) · 1283-1286
DOI: 10.1002/sdtp.70003_342 ↗Kurz gefasst
Fachartikel in der Zeitschrift SID Symposium Digest of Technical Papers, online erschienen am 11.09.2026. Verfasst von YI ZHOU und 1 weiteren Autor:innen. Die Arbeit nennt 6 Literaturangaben.
Abstract (Originalsprache)
To address the challenges of positioning accuracy degradation and power consumption constraints faced by Augmented Reality (AR) glasses under low‐light conditions, this paper proposes a low‐power visual Simultaneous Localization and Mapping approach, GSL‐SLAM (GCNv2‐SuperGlue‐LK‐SLAM), based on an adaptive dual‐mode framework.Under strict power constraints, AR glasses are required to achieve high‐precision and robust localization. However, traditional grayscale intensity‐based feature matching methods often fail in weak‐texture or illumination‐varying environments, while deep learning‐based approaches typically incur high computational costs, making them difficult to deploy on resource‐constrained wearable devices. GSL‐SLAM overcomes these limitations by integrating lightweight Lucas‐Kanade (LK) optical flow with deep learning features: efficient optical flow is employed for motion estimation during stable tracking, whereas the system automatically switches to a GCNv2‐based relocalization mode when tracking degrades or under challenging visual conditions.To further improve pose estimation accuracy, this paper introduces a feature matching strategy that integrates GCNv2 and SuperGlue, significantly increasing the number of reliable feature correspondences while maintaining real‐time performance. Experimental results demonstrate that the proposed GSL‐SLAM achieves a 18.8% improvement in positioning accuracy.
Zitieren
APA
ZHOU, Y., & JI, Y. (2026). P‐4.27: Towards Low‐Power Image Feature Extraction and Matching on AR Glasses. SID Symposium Digest of Technical Papers, 57(S1), 1283-1286. https://doi.org/10.1002/sdtp.70003_342
Vancouver
ZHOU Y, JI Y. P‐4.27: Towards Low‐Power Image Feature Extraction and Matching on AR Glasses. SID Symposium Digest of Technical Papers. 2026;57(S1):1283-1286. doi:10.1002/sdtp.70003_342
BibTeX
@article{ZHOU2026,
author = {ZHOU, YI and JI, YUAN},
title = {P‐4.27: Towards Low‐Power Image Feature Extraction and Matching on AR Glasses},
journal = {SID Symposium Digest of Technical Papers},
year = {2026},
volume = {57},
number = {S1},
pages = {1283--1286},
doi = {10.1002/sdtp.70003_342}
}
EUVIMED