LSC-54: A landmark-based dataset for colombian sign language

Area of Science:

• Computer Science
• Linguistics
• Human-Computer Interaction

Background:

• Global increase in hearing disabilities necessitates enhanced support for the deaf community.
• Limited access to sign language education, particularly Colombian Sign Language (CSL), poses a significant challenge.
• Computer Vision advancements offer potential for interactive sign language learning tools.

Purpose of the Study:

• To develop a time-series dataset for Colombian Sign Language (CSL).
• To create a technological tool to support CSL vocabulary acquisition.
• To address the need for accessible CSL educational resources.

Main Methods:

• Collected time-series data of static and dynamic CSL signs (colors, phrases, numbers) from 22 participants.
• Utilized 3D coordinate data of face, torso, and hands, processed using MediaPipe.
• Augmented the dataset by a factor of 45 to enhance diversity and model performance.

Main Results:

• A comprehensive dataset of Colombian Sign Language (CSL) signs was successfully created.
• The dataset includes 3D coordinate data for facial, torso, and hand movements.
• Data augmentation strategies were employed to improve dataset robustness.

Conclusions:

• The developed CSL dataset and associated processing pipeline can facilitate the creation of educational tools.
• This resource has the potential to improve access to Colombian Sign Language learning.
• Further development can leverage this dataset for advanced CSL recognition technologies.