Visão computacional para o reconhecimento de sinais do afabeto da língua de sinais Moçambicana

Abstract

Mozambican Sign Language (LSM) is the primary means of communication for the deaf community in Mozambique and plays a key role in social and educational inclusion. Despite its institutional recognition, the teaching and learning of LSM still face significant challenges, particularly the lack of digital educational resources and technological tools that support autonomous learning. In this context, this work proposes the development of an educational web application based on Computer Vision, aimed at the automatic recognition of the static alphabet of LSM as support for the initial learning process. The proposed approach relies on supervised machine learning, using data captured from a single video camera and hand feature extraction through landmarks. Due to the absence of a public and sufficiently comprehensive image dataset for LSM, the model was trained using a reference dataset of the American Sign Language (ASL) manual alphabet, taking advantage of the structural similarity observed between the static hand configurations of both sign languages. The study adopts a mixed-methods approach (qualitative–quantitative), combining contextual and linguistic analysis with an experimental evaluation of the model’s performance, using metrics such as accuracy, precision, recall, and F1-score. The results demonstrate the technical feasibility of the proposed approach for static gesture recognition, highlighting the potential of computer vision techniques to support the teaching and learning of LSM. The developed system represents an experimental prototype, with acknowledged limitations, but constitutes a relevant contribution to future research and to the development of technological solutions adapted to the Mozambican context.