VisionAid: A YOLO-Based Mobile Assistive System for Indoor Object Localization for the Visually Impaired
Aarav Jain
International Academy, Bloomfield Hills, Michigan, United States
Publication date: July 19, 2026
International Academy, Bloomfield Hills, Michigan, United States
Publication date: July 19, 2026
DOI: http://doi.org/10.34614/JIYRC2026I47
ABSTRACT
VisionAid is an application that is designed to assist visually impaired individuals in locating common objects using real-time object detection and auditory guidance. The aim of this research was to develop a YOLO-based object detection system and evaluate how capable it is of identifying seven everyday items: backpacks, glasses, headphones, keys, smartwatches, TV remotes, and wallets. A custom dataset consisting of 7,029 images was constructed using publicly available images, original photographs, and AI-generated images. The three different sources were used to better represent realistic room conditions, the scale of objects, and the distances between the camera and the object. The YOLO11n model was trained for 40 epochs and then evaluated on the test split. The model achieved an overall precision of 0.935, recall of 0.790, mAP50 of 0.877, and mAP50-95 of 0.701. These findings illustrate the strong potential VisionAid has in helping visually impaired individuals.
VisionAid is an application that is designed to assist visually impaired individuals in locating common objects using real-time object detection and auditory guidance. The aim of this research was to develop a YOLO-based object detection system and evaluate how capable it is of identifying seven everyday items: backpacks, glasses, headphones, keys, smartwatches, TV remotes, and wallets. A custom dataset consisting of 7,029 images was constructed using publicly available images, original photographs, and AI-generated images. The three different sources were used to better represent realistic room conditions, the scale of objects, and the distances between the camera and the object. The YOLO11n model was trained for 40 epochs and then evaluated on the test split. The model achieved an overall precision of 0.935, recall of 0.790, mAP50 of 0.877, and mAP50-95 of 0.701. These findings illustrate the strong potential VisionAid has in helping visually impaired individuals.
