This research addresses the accessibility barriers faced by visually impaired individuals when playing online board games, which are frequently designed exclusively for sighted users. The primary objective is to develop an assistive interface that acts as an intermediary layer between existing online gaming platforms and the user. By utilizing machine learning for real-time object recognition and providing corresponding textual or audio feedback, the system enables players to explore game boards and understand the placement of pieces, thereby fostering greater independence and social inclusion in digital gaming environments.
The study focuses on the abstract strategy game Quantik as a case study. The methodology involves training a convolutional neural network to classify game pieces by shape—specifically balls, cones, cubes, and cylinders—and color. The model was trained on a dataset of 500 images per class, achieving a validation accuracy of 94.5%. The interface, implemented in MATLAB, tracks the user's cursor or finger movements across the screen. When the user points to a specific area, the system identifies the piece and triggers a distinct audio cue, such as a gong or train sound, with pitch variations to distinguish between the player’s own pieces and those of an opponent.
The findings demonstrate that machine learning can effectively bridge the gap between visual game interfaces and non-visual interaction methods. By providing a sonified representation of the board, the interface allows users to construct a mental map of the game state. Future development aims to improve classifier accuracy, refine the sonification process to be more intuitive, and reduce the reliance on sighted individuals for initial dataset creation. The researchers suggest that this framework is scalable and could be adapted for other board or card games, ultimately enhancing the quality of life and personal fulfillment for visually impaired players.
