AI-Powered Smart Glasses Using the ESP32 Processor

Abstract

Vision impairment affects over 2.2 billion people worldwide and continues to impose severe barriers to independent daily life, from safe outdoor navigation to reading signage and recognising familiar faces. Conventional aids such as the white cane provide limited scene awareness and cannot describe objects, read text, or interpret traffic signals. The convergence of affordable edge-computing hardware, mature deep learning frameworks, and free-tier cloud AI services has spawned a new generation of AI-powered smart glasses that begin to close this gap. This paper presents a structured comparative review of five peer-reviewed conference publications spanning 2020 to 2025, each proposing a working smart-glasses prototype for blind and visually impaired users. The systems surveyed range from a minimal ESP32-based image-relay prototype to a classical Viola-Jones OpenCV pipeline, a YOLOv3 client-server deep learning design, a YOLOv5 and Kalman-filter sensor-fusion wearable, and a Google Coral Edge TPU driven autonomous navigation system. Each design is analysed across hardware platform, detection algorithm, sensing modalities, user feedback channel, and quantitative performance where available. A cross-paper comparison is supported by a generalised four-layer architectural model, a taxonomy of detection algorithms, and a hardware evolution timeline. The survey concludes with a discussion of open challenges and concrete future directions including large language model integration and hybrid cloud-edge processing.