LSC-54: A Landmark-Based Dataset for Colombian Sign Language

According to the World Health Organization, hearing disabilities are increasing globally, highlighting the need to support the deaf community. In countries such as Colombia, access to sign language education remains limited. Recent advances in Computer Vision have proven effective in promoting sign language learning through real-time interaction. In response to this need, a time-series dataset was developed to represent signs from Colombian Sign Language (CSL), including 11 colors, 33 courtesy phrases, and the numbers 1–10. This dataset served as input for the development of a technological tool aimed at supporting the teaching of CSL vocabulary. It includes both static and dynamic signs, represented as 3D coordinate data of the face, torso and hands. The dataset was built using recordings from 22 participants, each performing sign between three to five times. To increase data diversity and improve model performance, the dataset was augmented by a factor of 45. The data acquisition process involved: A user interface for recording sign language videos; post-processing (extracting coordinates via MediaPipe, and imputing missing hand data), and storing processed data in JSON files.