Libras

LIBRAS, acronym of the Portuguese name "Lingua BRAsileira de Sinais", is the oficial brazilian sign language. The dataset contains 15 classes of 24 instances each, where each class references to a hand movement type in LIBRAS. The hand movement is represented as a bi-dimensional curve performed by the hand in a period of time. The curves were obtained from videos of hand movements, with the Libras performance from 4 different people, during 2 sessions. Each video corresponds to only one hand movement and has about 7 seconds. In the video pre-processing, a time normalization is carried out selecting 45 frames from each video, in according to an uniform distribution. In each frame, the centroid pixels of the segmented objects (the hand) are found, which compose the discrete version of the curve F with 45 points. All curves are normalized in the unitary space. In order to prepare these movements to be analysed by algorithms, we have carried out a mapping operation, that is, each curve F is mapped in a representation with 90 features, with representing the coordinates of movement. Each instance represents 45 points on a bi-dimensional space, which can be plotted in an ordered way (from 1 through 45 as the X co-ordinate) in order to draw the path of the movement.

This is a pre-processed version of the dataset saved in numpy format. The original dataset is obtained from UCI.

The data are 3-dimensional arrays of shape [n_samples, time_steps, n_variables]. The data can be loaded as follows:

loaded_data = np.load("LIB.npz")
Xtr = loaded_data['Xtr'] # Training data of shape (180, 45, 2)
Ytr = loaded_data['Ytr'] # Training labels of shape (180, 1)
Xte = loaded_data['Xte'] # Test data of shape (180, 45, 2)
Yte = loaded_data['Yte'] # Test labels of shape (180, 1)