Mexican Sign Language Alphabet (dynamic signs only)

Overview

The Dynamic Alphabet dataset for Mexican Sign Language (LSM) is a comprehensive and diverse collection of the signs J, K, Ñ, Q, X, and Z, in order to complement LSM static alphabet https://doi.org/10.5281/zenodo.10067509. This dataset is designed to facilitate research and the development of tools focused on both static and dynamic sign recognition.

Data Representation

• The dataset consists of videos of a hand performing the sign, captured in a controlled environment with good lighting and a uniform green background.
• Since these signs are dynamic, they are represented by videos that allow visualization of how the sign evolves over time.
• In total, data were collected from 20 participants, ensuring a wide variety of sign representations within this set.

Data Collection

• Data were gathered in a controlled environment. The subject stands in a fixed position and extends their hand sufficiently far from the body to obtain a clearer view of the sign.
• Two views respect to the camera were recorded: (i) frontal, and (ii) 45°-right-rotated perspective.
• As mentioned previously, these signs were recorded against a uniform green background, ensuring minimal loss in detection.
• Each participant performed each sign five times.

Data Partitioning

The dataset is divided into two subfolders, "train" and "test":

• The "train" folder contains data from the first 18 participants.
• The "test" folder contains data from the remaining 2 participants.

• Each video has the following naming convention: S<SubjectId>-<Letter>-<View>-<RepetitionNumber>.