Advancing AI-based Assistive Systems for Visually Impaired People: Multi-Class Object Detection and Currency Classification

Blind and Visually Impaired (BVI) persons depend on the assistive systems. An intelligent assistive system may provide situational awareness and support in the age of machine learning and computer vision with the help of deep learning and computer vision. Among many, multi-classed object detection and currency classification are the two key means for providing situational awareness and environmental information to the BVI people. Thus, the objective of this research is to propose a modified YOLOv7 and a YOLOv8 algorithm for multi-classed object detection and currency classification, respectively. To attain this objective, a modified YOLOv7 and a YOLOv8 was proposed and evaluated considering the datasets for object identification and currency classification, respectively. The dataset for object detection was created from scratch that includes 3046 photos and spans 20 distinct classes, while the dataset used for currency classification used 295 photos divided into eight distinct groups. The evaluation results showed that in comparison to the basic YOLOv7 (89.3% mAP) the modified YOLOv7 has obtained 91.4% mAP for multi-classed object detection, while the YOLOv8 has reached 94.6% mAP for currency classification comparing to the Radial Basis Function Network (91.51% mAP).