TinyML-Powered Handwritten Digit Recognition Device for the Visually Impaired

Accessing information in an easily understandable format remains a significant challenge for visually impaired individuals. Conventional handwritten digit recognition systems often rely on computationally intensive models, limiting their deployment on portable, low-cost devices. This paper presents a TinyML-based system for real-time handwritten digit recognition designed specifically to assist visually impaired users. Leveraging a Convolutional Neural Network (CNN) deployed on a Raspberry Pi, the system delivers accurate digit recognition with both visual and audio feedback, enabling independent identification of hand-written digits encountered in everyday activities. The integrated solution combines a Pi camera module, a 3.5-inch display, and an audio speaker, resulting in a compact, portable, and user-friendly device. The methodology involves training the CNN on a curated dataset and optimizing it for edge deployment using TinyML techniques, ensuring low-latency and energy-efficient operation. Experimental results demonstrate the system’s capability for efficient and reliable digit recognition, highlighting its potential to enhance accessibility and empower visually impaired individuals through affordable, real-time assistive technology. Keywords - Handwritten digits recognition, Convolutional Neural Network (CNN), Visual impairment, Raspberry Pi, TinyML, Assistive technology development