Hand Gesture Accelerometer and Gyroscope Dataset (HGAG-DATA)
Description
This dataset includes information from 43 healthy participants (26 males and 17 females). The participants encompass an extensive age range of 18 to 69 years and are classified by dominant hand, comprising 34 right-handed and 9 left-handed individuals. Furthermore, they are categorized according to physical activity levels, comprising 28 non-athletic individuals and 15 athletic individuals. All participants were equipped, trained, and directed to perform 11 essential gestures relevant in many circumstances and closely linked to daily life requirements. The motions encompass clapping, coin flipping, finger snapping, fist making, horizontal wrist extension, index finger flicking, index thumb tapping, shooting, thumbs up, wrist extension, and wrist flexion. The dataset comprises 23,650 six-dimensional gesture samples, captured at a rate of 550 samples per subject following 50 repetitions of each of the eleven motions (11x50x43 = 23,650). Each gesture sample consisted of six directional time series signals, incorporating two signals for each of the x, y, and z axes from the accelerometer and gyroscope sensors. This dataset comprises a total of 141,900 signals (23,650 x 6 = 141,900). Consequently, this can serve as a significant resource for handling extensive data sets, such as those utilized in the development of machine learning and deep learning models, and as a reference dataset, it facilitates model benchmarking. The data will be especially helpful for societies that work with biomedical signals when they are trying to recognize and classify hand gestures for use in human-computer interaction (HCI) tasks. The dataset has been uploaded and is accessible online in two distinct hierarchical configurations. The initial structure categorizes the data according to the gesture name, but the subsequent structure arranges it by subject number. This configuration enhances potential advantages and facilitates management, permitting the reassembly of data in many formats as required.
Files
Steps to reproduce
Before commencing the signal recording for the motions being studied, each participant was posed a series of questions. The aim of these enquiries was to gather background information on each participant and conduct a demographic assessment. This was conducted to identify the appropriate individuals for data collection. The "HGAG-DATA" dataset has been acquired by a new human-computer interaction (HCI) technology. The data measurements were acquired using the tri-axial accelerometer and tri-axial gyroscope of the MPU-92/65 sensor module. This module is intended to be worn as a watch, enabling smooth and seamless operation. This module is connected to the basic ESP32 DEVKIT V1 microcontroller, which operates as a transmitter, sending samples of gesture data to the receiver via Wi-Fi. The receiver is the second ESP32 microcontroller of identical specifications, directly connected to the laptop via a USB cable. The six-axis raw data for each performed gesture were collected at a sampling frequency of 200 Hz by the primary ESP32 microcontroller, which transmits the data to the laptop via the secondary ESP32 microcontroller over a Wi-Fi connection. The data undergoes additional processing via a 4th order Butterworth bandpass filter with a frequency range of 30-90 Hz to eliminate motion artefacts and high-frequency noise. This enabled the researchers to concentrate exclusively on the noiseless signals produced by the muscles. It is crucial to acknowledge that the filter applied to the raw data was selected following extensive testing and modifications to its kind, frequency, and degree.