毫米波多场景上肢行为识别数据集

使用毫米波（mmWave）雷达的“空中”手势识别及其在智能家居的自然人机交互中的应用显示了其潜力。然而，最先进的作品在有限的手势空间、易受周围干扰和离线识别方面仍然不足。我们招募了25名志愿者，其中包括11名男性和14名女性，身高157至183厘米，体重46至93公斤，年龄19至76岁，预先定义了10个多关节手臂姿势。具体而言，我们收集了13个锚位置的数据。他们被要求执行每个预定义的手势30次。我们从7个室内场景收集训练手势数据，例如卧室、书房和客厅。训练数据集在97天内存档，总共累积了22000多个样本，这消除了志愿者在做相同手势时的僵硬。

Air gesture recognition using millimeter-wave (mmWave) radar and its application in natural human-computer interaction for smart homes have demonstrated significant potential. However, existing state-of-the-art works still suffer from limitations including restricted gesture space, susceptibility to ambient interference, and exclusive reliance on offline recognition. We recruited 25 volunteers, comprising 11 males and 14 females, with heights spanning 157 cm to 183 cm, weights ranging from 46 kg to 93 kg, and ages between 19 and 76 years old. Ten predefined multi-joint arm gestures were designed in advance. Specifically, we collected data from 13 anchor positions. All volunteers were instructed to perform each predefined gesture 30 times. We gathered training gesture data across 7 indoor scenarios, such as bedrooms, home offices, and living rooms. The training dataset was accumulated over a period of 97 days, yielding a total of over 22,000 samples, which mitigated the stiffness that volunteers exhibited when repeatedly performing the same gesture.

搜集汇总

背景与挑战

背景概述

该数据集是一个专注于毫米波雷达的上肢行为识别数据集，用于智能家居人机交互研究。它包含25名志愿者在7个室内场景中执行的10个多关节手臂姿势数据，累计超过22000个样本，旨在通过多场景和多志愿者设计提高识别模型的泛化能力和鲁棒性。数据量551.99MB，共1159个文件，发布于2023年，适用于计算机科学技术领域的应用开发。