Hand Gesture Accelerometer and Gyroscope Dataset (HGAG-DATA)

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.

本数据集涵盖43名健康受试者的相关信息，其中男性26名，女性17名。受试者年龄跨度广泛，为18至69岁，并按利手（dominant hand）类型分类：34名为右利手，9名为左利手。此外，受试者还根据身体活动水平划分为两类：28名非运动人群，15名运动人群。
所有受试者均完成设备佩戴培训，并被要求完成11种在多种场景中常用、且与日常生活需求紧密相关的典型手势动作，具体包括：拍手、抛硬币、弹指、握拳、腕部水平伸展、食指弹动、食指拇指轻叩、射击手势、竖大拇指、腕部伸展以及腕部屈曲。
本数据集共包含23650个六维手势样本，采集规则为每名受试者完成11种手势各50次重复，以550样本/受试者的速率进行采集（计算公式为11×50×43=23650）。每个手势样本包含六维方向时间序列信号，分别来自加速度计（accelerometer）与陀螺仪（gyroscope）传感器的x、y、z三轴，每轴对应2个信号。数据集总计包含141900个信号（23650×6=141900）。
该数据集可作为处理大规模数据集的重要资源，例如用于机器学习与深度学习模型的开发；同时作为基准参考数据集，可助力模型性能基准测试。对于从事生物医学信号相关研究、旨在识别人手手势以应用于人机交互（Human-Computer Interaction, HCI）任务的科研团队而言，本数据集尤为实用。
本数据集已上传至网络，提供两种不同的层级组织格式可供访问：第一种格式按手势名称对数据进行分类，第二种格式则按受试者编号进行排序。该设计不仅提升了数据的潜在应用优势，还便于管理，支持根据需求以多种格式重新组合数据。