DU-ASL-DATA-GLOVE-DB
Hardware
The primary hardware is a data glove consisting of three units, namely sensing, processing, and onboard power regulation unit. The sensing unit is comprised of five 2.2" flex sensors (SEN-10264) and an IMU (MPU-6050) which has a triaxial accelerometer and a triaxial gyroscope.
Dataset
Overview. We explored 40 signs from the standard ASL dictionary that including 26 letters and 14 words. Among these signs, 24 require only a certain finger flexion and no hand motion; hence, are addressed as static signs or gestures. Conversely, the remaining 16 signs need hand motion alongside the finger flexion to portray meaningful expression according to the ASL dictionary. Moreover, we collected the signs from 25 subjects (19 Male and 6 Female) in separate data recording sessions with a consistent protocol. Overall, three channels for acceleration in both body and earth axis, three for angular velocity, four for quaternion, and five for flex sensors were recorded in the dataset.
The data was recorded by the data glove processing unit which was connected to a laptop for data storage via USB. The sampling frequency is set to 100 Hz and each gesture was repeated 10 times to record the performance variabilities of each subject. However, during a few sessions denoted in the dataset supplementary information, the laptop charger was connected which resulted in AC-induced noise all over those specific recorded data.
Data recording protocol. Before starting the recording process, each subject signed an approval form for the usage of their data in this research and was briefed about the data recording steps. As the subjects were not familiar with the signs before the study, they were taught each sign before the data recording via online video materials. The data was recorded by the data glove and stored on the laptop at the same time. Hence, a Python script was used on the laptop to make the handshake between the two devices and to store the data in separate folders as per the signs and the subjects.
At the beginning of each data recording session, the subjects were prompted to declare their subject id and the gesture name. Afterward, a five-second countdown is prompted on the laptop screen for preparation. Each instance of the gesture data is recorded for a 1.5 seconds window and the subjects can easily perform their gesture once within that window. In a single gesture recording session, this process is repeated 10 times.
