NAVISIGHT: A Deep Learning and Voice-Assisted System for Intelligent Indoor Navigation of the Visually Impaired

Mrs. V.Vijaya Lakshmi 1, Karapakula Balaji Sai Vrushadree2, K.Eekshitha3, M.Hemavathi4, T.Gurubunny5

1 Assistant Professor, Department of Electronics & Communication Engineering , Annamacharya Institute Of Technology &Sciences, Tirupati, Andhra Pradesh, India, vijji.siri1992@gmail.com

2 U.G Scholar, Department of Electronics & Communication Engineering , Annamacharya Institute Of Technology & Sciences, Tirupati, Andhra Pradesh, India, kbsv2004@gmail.com

3 U.G Scholar, Department of Electronics & Communication Engineering , Annamacharya Institute Of Technology & Sciences, Tirupati, Andhra Pradesh, India,koppalaeekshitha@gmail.com

4 U.G Scholar, Department of Electronics & Communication Engineering , Annamacharya Institute Of Technology & Sciences, Tirupati, Andhra Pradesh, India, hemameduri320@gmail.com

5 U.G Scholar, Department of Electronics & Communication Engineering , Annamacharya Institute Of Technology & Sciences, Tirupati, Andhra Pradesh, India, bunnyt903@gmail.com

Abstract – Autonomous indoor navigation for visually impaired users is constrained by the limited contextual awareness of conventional aids such as white canes and guide dogs. This paper presents an embedded AI navigation system that integrates YOLO-based object detection, semantic room classification, ultrasonic proximity sensing, and IoT emergency alerting into a unified, standalone framework. Detected objects drive real-time room inference, while sensor fusion between ultrasonic range data and visual outputs enhances obstacle localisation. Synthesised voice prompts deliver context- aware guidance without dependency on external processing . Experimental evaluation confirms high detection accuracy, low-latency performance, and practical deployability in diverse indoor environments.

Keywords: Assistive technology, computer vision, object detection, room classification, deep learning, IoT, sensor fusion