UT-Kinect (UTKinect-Action3D Dataset)

UT-Kinect 数据集是用于从深度序列中识别动作的数据集。这些视频是使用单个固定 Kinect 捕获的。有10种动作类型：走路、坐下、站起来、捡起、搬运、扔、推、拉、挥手、拍手。有10个科目，每个科目执行两次每个动作。记录了三个通道：RGB、深度和骨骼关节位置。三通道同步。帧率为 30f/s。

The UT-Kinect Dataset is a benchmark dataset designed for action recognition from depth sequences. Videos were captured using a single fixed Kinect sensor. It includes 10 action categories: walking, sitting down, standing up, picking up, carrying, throwing, pushing, pulling, waving, and clapping. There are 10 subjects in total, with each subject performing each action twice. Three synchronized data channels are recorded: RGB, depth maps, and skeletal joint positions, with a frame rate of 30 fps.

搜集汇总

背景与挑战

背景概述

UT-Kinect数据集是一个用于动作识别的多模态数据集，专注于从深度序列中识别10种常见人类动作，如走路、坐下和挥手。它包含10个受试者执行每个动作两次的视频数据，提供同步的RGB、深度和骨骼关节位置三个通道，帧率为30fps，适用于计算机视觉和动作识别研究。该数据集由德克萨斯大学奥斯汀分校于2012年发布，旨在支持视图不变的动作识别算法开发。