Computer-Vision- and Edge-Enabled Real-Time Assistance Framework for Visually Impaired Persons with LPWAN Emergency...

Area of Science:

• Assistive Technology
• Computer Vision
• Edge Computing

Background:

• Existing assistive technologies for the visually impaired often lack real-time, universal capabilities in dynamic environments.
• Current systems face trade-offs in range, latency, or reliability in varied indoor/outdoor settings.
• A need exists for integrated solutions combining object detection, communication, and emergency signaling.

Purpose of the Study:

• To propose a comprehensive assistive framework for visually impaired individuals.
• To integrate computer vision, edge computing, and a dual-channel communication architecture.
• To enhance independent mobility and quality of life through real-time assistance.

Main Methods:

• Utilized the YOLOv5 deep-learning model for high-performance object detection (precision 0.988, recall 0.969, mAP 0.985).
• Implemented edge computing for fast, local processing and decision-making, offloading central servers.
• Designed a dual-channel communication system: Wi-Fi primary link and LoRaWAN (Low-Power Wide-Area Network) for emergency alerts.
• Incorporated an IoT-based panic button for location-tagged emergency transmissions.

Main Results:

• Demonstrated low latency and reliable system operation in diverse real-world conditions.
• Achieved high accuracy in detecting obstacles, paths, and assistive tools.
• The system proved effective in dynamic indoor and outdoor environments.
• Validated the cost-effectiveness and scalability of the proposed architecture.

Conclusions:

• The developed framework significantly enhances real-time assistance for the visually impaired.
• The integration of YOLOv5, edge computing, and dual communication offers a robust solution.
• The system has strong potential to improve independent mobility and safety for visually impaired users.
• The architecture is suitable for deployment in complex environments requiring immediate support.