MINDS-Libras Dataset (RGB-D sensor data)

Brazilian Sign Language (Libras) data set with 20 signs for sign language and gesture recognition benchmark:

- Acontecer (To happen)
- Aluno (Student)
- Amarelo (Yellow)
- América (America)
- Aproveitar (To enjoy)
- Bala (Candy)
- Banco (Bank)
- Banheiro (Bathroom)
- Barulho (Noise)
- Cinco (Five)
- Conhecer (To know)
- Espelho (Mirror)
- Esquina (Corner)
- Filho (Son)
- Maçã (Apple)
- Medo (Fear)
- Ruim (Bad)
- Sapo (Frog)
- Vacina (Vaccine)
- Vontade (Will)

Each one of the signs was recorded 5 times by 12 signers, using a Chroma Key background. Among the signers are men and women with basic to advanced knowledge in Libras. 

The RGB-D sensor (kinect v2) available the RGB videos (1920 x 1080) and depth videos (640 x 480) in "mp4" format, and the body points and face data in "txt" file.

The body file has seven different information (Position X, Y and Z; Orientation X, Y and Z; TrackingState; LeftHandState; RightHandState; ColorPosition X and Y; and DepthPosition X and Y) about the 25 points: (1) Spine Base, (2) Spine Mid, (3) Neck, (4) Head, (5) Shoulder Left, (6) Elbow Left, (7) Wrist Left, (8) Hand Left, (9) Shoulder Right, (10) Elbow Right, (11) Wrist Right, (12) Hand Right, (13) Hip Left, (14) Knee Left, (15) Ankle Left, (16) Foot Left, (17) Hip Right, (18) Knee Right, (19) Ankle Right, (20) Foot Right, (21) Spine Shoulder, (22) Hand Tip Left, (23) Thumb Left, (24) Hand Tip Right and (25) Thumb Right. There are 13 lines (or data) for each frame. This order is repeated sequentially up to 1950 lines (13 lines $\times$ 150 frames), representing the sign video.

Regarding to the face data, the same organisation was adopted. In this case, we have seven information (FaceBox, FaceRotation, HeadPivot, AnimationUnit, FaceModel X, Y and Z; ColorFaceModel X and Y; and DepthFaceModel X and Y), describing 11 data, distributed in 1650 (11 lines $\times$ 150 frames) lines in the ``.txt'' file.

(Former name: Libras-20)