Building an AI-Integrated Braille Recognition Model for Blind Education: From Mechanical Devices to Personalized Learning Systems

The context of inclusive education for blind people lacks suitable learning materials and traditional devices are difficult to expand, creating a need for a personalized technological solution. This study proposes and evaluates an AI-integrated Braille recognition and conversion model, designed in a three-layer architecture including terminals, AI engines, and learning management systems. At the device layer, Raspberry Pi 5 connects a camera placed near the paper surface with a choice of OV9281 or IMX708 sensors, displayed via a vibrating dot matrix and speakers. At the AI layer, the processing chain applies region of interest cropping, local contrast enhancement, noise reduction, adaptive thresholding, skew and distortion correction, then segmenting lines, words, characters, recognizing them using optical and deep learning models such as CRNN or DBNet (for optical character recognition), and mapping the results to a two-level intensity vibration rule, extended to shapes and twelve basic colors. In the learning management layer, interactive data is used for adaptive suggestions using lightweight algorithms such as nearest neighbor and LightGBM (for generating personalized learning suggestions). Experiments with 15 blind students in Da Nang show recognition accuracy greater than 90% on the basic character set, supporting text reading, basic math and geometric perception. The results confirm the feasibility of the integration direction and suggest prospects for wide deployment in inclusive education, while also suggesting development directions for data expansion, hardware optimization and assessment standardization.