Banknote and obstacle detection system for visually impaired people

This paper presents the development of two perception modules for visually impaired people. In the first module, a robust YOLO-based neural network model is proposed to recognise the denomination of American, European, Mexican, and Colombian banknotes, achieving a detection performance over 97 % over all denominations. In the second module, an efficient obstacle detection algorithm based on stere-ovision is developed to prevent collisions while walking. Several performance tests and simulations were carried out using each module, proving that each is accurate enough to help blind people to complete the proposed perception tasks.