Varying-view RGB-D Action Dataset
收藏资源简介:
Varying-view RGB-D Action Dataset是由电子科技大学未来媒体研究中心创建的大型数据集,旨在支持任意视角的人体动作识别研究。该数据集包含从8个固定视角和360度全视角捕获的RGB视频、深度图像和骨骼序列,总计邀请了118人执行40种动作类别。数据集不仅视角多样,样本数量庞大,而且是首个包含完整360度视角序列的数据集,适用于多视角、跨视角及任意视角的动作分析。此外,数据集还应用于健身辅助机器人等领域,为解决视角变化带来的动作识别挑战提供了丰富的数据资源。
The Varying-view RGB-D Action Dataset is a large-scale dataset created by the Future Media Research Center of the University of Electronic Science and Technology of China, aiming to support research on human action recognition from arbitrary viewpoints. It includes RGB videos, depth images and skeleton sequences captured from 8 fixed viewpoints and a full 360-degree panoramic view. A total of 118 participants were invited to perform 40 distinct action categories. This dataset not only features diverse viewpoints and a massive number of samples, but is also the first dataset that contains complete 360-degree viewpoint sequences, which is applicable to multi-view, cross-view and arbitrary-view action analysis. Furthermore, the dataset has been applied in fields such as fitness assistance robotics, providing abundant data resources for addressing the challenges of action recognition caused by viewpoint variations.

- 1A Large-scale Varying-view RGB-D Action Dataset for Arbitrary-view Human Action Recognition电子科技大学未来媒体研究中心 · 2019年



