Aplicação de redes neurais na detecção de símbolos da Língua Brasileira de Sinais (Libras)

• Autor(a) principal: Zhang, Ana Carolina
• Data de Publicação: 2024
• Outros Autores: Rodrigues, Heloisa Mariani
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da PUC_SP
• Texto Completo: https://repositorio.pucsp.br/jspui/handle/handle/42330
• topic: CNPQ::ENGENHARIAS; Língua Brasileira de Sinais (Libras); Detecção de sinais; Rede Neural Convolucional (CNN); MediaPipe; Acessibilidade; Tecnologia assistive; Comunidade surda; Inclusão; Reconhecimento de gestos; ROBOFLOW; MLP (Multilayer Perceptron); Brazilian Sign Language (Libras); Sign detection; Convolutional Neural Network (CNN); MediaPipe; Accessibility; Assistive technology; Deaf community; Inclusion; Gesture recognition; ROBOFLOW; MLP (Multilayer Perceptron)

In this study, we explored strategies to detect Brazilian Sign Language (Libras) signs, facing significant challenges. Initially, using a Convolutional Neural Network (CNN) with the UEFS dataset, we encountered inaccuracies due to the low resolution of the images. When applying the CNN to the ROBOFLOW dataset, the results were still unsatisfactory, indicating the complexity of the problem and the presence of overfitting. We chose to integrate MediaPipe with the ROBOFLOW data, achieving an accuracy of 93.66%. The results underscore the importance of technological innovation in promoting inclusion and accessibility for the deaf community, with the potential to significantly improve the quality of life and inclusion of deaf individuals