A Deep Learning–Driven Assistive Vision Framework for the Visually Impaired using YOLOv5 and Pyttsx3

View at publisher Summarise with VIRAT
Authors: Praveena R, Madhan S, Navinkumar R, Chandra Bose V
Cited by: 0
References: 10

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Visual impairment severely influences mobility, spatial cognition, and autonomy. As a result, there is a growing demand for intelligent assistive systems capable of understanding complex environments. This paper presents a general-purpose, offline assistive vision system designed to enhance the experience of partially sighted users by providing real-time object detection, relative distance estimation, and scene recognition through naturalistic auditory feedback. The primary objective of this work is to develop a fast, portable, and network-free solution that increases environmental awareness during navigation while reducing cognitive load. The proposed system integrates three deep learning components within a unified processing pipeline: YOLOv5s for object detection, a bounding-box–based approximation model for distance estimation, and EfficientNet-B0 trained on the Places365 dataset for contextual scene classification. Video frames are processed in real time using PyTorch and OpenCV, while detected information is converted into spoken descriptions via the pyttsx3 text-to-speech engine. All predictions are overlaid onto the live video stream while simultaneous auditory feedback is provided. Experimental evaluations conducted in both indoor and outdoor environments demonstrate real-time performance of 2428 FPS without reliance on cloud services. Object detection remained consistent across classes, distance estimation exhibited low average error sufficient for mobility guidance, and scene classification remained stable even under rapid camera motion. Overall, the combined outputs deliver reliable and informative feedback, indicating that the proposed system is effective for use in practical assistive navigation devices.

Fields of Study & Subjects

Assistive Technology in Communication and Mobility Advanced Neural Network Applications Tactile and Sensory Interactions

Publication details

Journal / source
International Conference on Computing, Communication, Control and Cyber-Physical Systems (I5CPS)
Pages
1–6
Publication date
26 February 2026
Language
EN