动态手势识别多模态数据

本数据集面向动态手势识别任务，构建了包含图像模态与时序模态的多模态样本库。图像模态包括每个手势动作的多帧二值化手部图像，时序模态为每帧图像的时间信息。数据采集于2023年6月至2024年3月期间，在清华大学智能技术与系统国家重点实验室使用高清工业相机拍摄，采集过程中保持光照稳定、背景干净，确保手势动作完整连续。图像分辨率统一为640×470，图像数据以.jpg格式保存，标签信息以.txt格式保存。每个手势样本经过多人多次采集与人工审核标注，确保数据一致性。通过标准化处理、重复采集一致性测试等手段进行质量控制，标准差控制在±5%以内。该数据集为动态手势识别模型提供训练和测试支持，具有良好的重用价值，适用于人机交互、辅助控制等领域的研究开发。

This dataset is developed for dynamic gesture recognition tasks, and constitutes a multimodal sample library encompassing both image and temporal modalities. The image modality includes multi-frame binarized hand images for each gesture action, while the temporal modality stores the timestamp information corresponding to each image frame. The data was collected from June 2023 to March 2024 using high-definition industrial cameras at the State Key Laboratory of Intelligent Technology and Systems, Tsinghua University. During the collection process, stable lighting and clean background were maintained to ensure complete and continuous gesture motions. All images have a uniform resolution of 640×470, with image data saved in .jpg format and label information stored in .txt format. Each gesture sample was collected multiple times by multiple participants and underwent manual review and annotation to ensure data consistency. Quality control was conducted via standardized processing, consistency tests of repeated collections and other means, with the standard deviation controlled within ±5%. This dataset provides training and testing support for dynamic gesture recognition models, boasts excellent reusability, and is applicable to research and development in fields such as human-computer interaction and auxiliary control.

搜集汇总

背景与挑战

背景概述

该数据集是一个面向动态手势识别的多模态样本库，包含图像模态（多帧二值化手部图像）和时序模态（时间信息），数据采集于清华大学实验室，经过严格质量控制（如标准化处理和一致性测试，标准差在±5%以内），确保手势动作完整连续。它适用于人机交互和辅助控制等领域的模型训练与测试，具有较高的重用价值。