Automated Object Identification for the Visually Impaired

In Bangladesh, there exists a significant population of visually impaired individuals, facing daily challenges in navigation, obstacle identification, financial transactions, and social interactions. While previous research has focused on specialized tasks such as object detection, recognition, and facial recognition, there is a lack of studies integrating these functionalities into a unified automated system. This paper presents a comprehensive assistive system that uses exclusive datasets and advanced machine learning models, including YOLOv5 and ResNet-50, to identify and classify objects, currencies and faces. The system aims to provide real-time information about nearby objects, various currency denominations, and recognized faces, thereby assisting visually impaired individuals in navigating their environment, conducting secure transactions, and improving social interactions. Each module of the system undergoes thorough evaluation through experimentation. The object detection module achieves a mean average precision (MaP) of 98.2% for all classes over 2188 photos, while the currency categorization module achieves a top-1 accuracy of 99.7% when tested on 374 pictures. The face recognition module can accurately recognize 97.6% of 128 face images.