EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Hybrid Edge-Cloud Wearable Framework for Context-Aware Spatial Navigation

Viriyala Sri Anima Padmini, Vithya Ganesan

Computer and Decision Making: An International Journal · 28.10.2026 · 3 (6S) · 2659-2682

DOI: 10.1922/comdem.v3i6s.19016 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Computer and Decision Making: An International Journal, online erschienen am 28.10.2026. Verfasst von Viriyala Sri Anima Padmini und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Student- This Independent indoor mobility is a key unsolved problem of visually impaired people: GPS-assisted location cannot be provided in closed areas in low-light circumstances. The proposed model with a hybrid edge-cloud wearable navigation system deployed on smart eyewear and combines three sensing modalities RGB vision, inertial measurement, and thermal imaging that are spread across a Raspberry Pi 4B edge node and a remote, fully equipped, graphical processing server.YOLOv11 object detects and classifies hazards at the floor level, Dense Prediction Transformer (DPT) domain adapted to detect objects estimates dense monocular depth, and GPT-4 generates context-aware spoken navigation directions, trained on a structured spatial prompt schema. An Extended Kalman Filter is a combination of MPU-6050 inertial data and monocular camera pose information in SE(3) to generates drift-reduced updates to the scene graph as the head moves continuously. MLX90640 thermal camera is homographic ally correlated to the RGB frame, and fused using luminance-weighted confidence fusion, and allows the person to be detected with high accuracy in zero-light conditions. The structured prompt schema represents the present scene graph, IMU-corrected user pose, thermal alert status, and navigation history into each query to GPT-4, hallucinating less than 1%. A levelled fallback system gracefully transitions between LLM response and haptic buzzer notifications through the use of a gradual fallback to cached scene graph output in cases where the network or latency limitations are surpassed. The system has an IMU-stabilized pose and thermal-enhanced person detection which translates to a 93.5 percent detection accuracy and 94.1 percent goal of navigation at an average mean end-to-end latency of 123 ms over 180 live indoor runs.

Zitieren

APA

Padmini, V. S. A., & Ganesan, V. (2026). Hybrid Edge-Cloud Wearable Framework for Context-Aware Spatial Navigation. Computer and Decision Making: An International Journal, 3(6S), 2659-2682. https://doi.org/10.1922/comdem.v3i6s.19016

Vancouver

Padmini VSA, Ganesan V. Hybrid Edge-Cloud Wearable Framework for Context-Aware Spatial Navigation. Computer and Decision Making: An International Journal. 2026;3(6S):2659-2682. doi:10.1922/comdem.v3i6s.19016

BibTeX
@article{Padmini2026,
  author = {Padmini, Viriyala Sri Anima and Ganesan, Vithya},
  title = {Hybrid Edge-Cloud Wearable Framework for Context-Aware Spatial Navigation},
  journal = {Computer and Decision Making: An International Journal},
  year = {2026},
  volume = {3},
  number = {6S},
  pages = {2659--2682},
  doi = {10.1922/comdem.v3i6s.19016}
}
Kostenlos starten →