Transfer Learning Based Indoor Object Recognition for Visually Impaired People

Visually impaired individuals encounter significant challenges when navigating and interacting with indoor environments. This paper proposes the development of an indoor object recognition system based on transfer learning, for visually impaired people. It transfers the learning of common object detection models pre-trained using ImageNet, rich dataset, to MyNursingHome dataset, which is a relatively limited dataset specifically created for indoor object detection and recognition. The results indicate that the YOLOv8 model achieves an average accuracy of 96%, demonstrating its suitability for implementation in the mobile application to assist visually impaired individuals.