Automatic Time-aware Recognition of Brazilian Sign Language Based on Dynamic Time Warping

The Brazilian Sign Language (Libras) is a crucial communication medium for the deaf community in Brazil, yet it poses significant challenges for recognition and translation tasks. This paper presents a novel approach using Fast Dynamic Time Warping (FastDTW) for recognizing Libras signs in video streams. This approach aims to bridge the communication gap between deaf and hearing individuals, enhancing accessibility and reducing social marginalization. The methodology leverages MediaPipe to extract key hand and body landmarks, which are then used to compute angular features for accurate sign recognition. Experiments were conducted on the MINDS-Libras dataset, and the results demonstrated a high recognition accuracy, outperforming traditional methods. Furthermore, when the proposed model is applied to the INCLUDE-50 dataset containing signs from a different sign language, it performs competitively without relying on deep learning techniques.