手势体势数据集

本数据集主要面向手势/体势动作行为识别及意图认知的研究及其需求建设。录制场地为浙江工业大学计算机视觉研究所实验室。采用三套设备共六个传感器共同采取人体动作，包括3个Intel RealSense和3个Event-based Camera。采集数据集所需传感器按如下方案摆放：以1个Intel RealSense和1个Event-based Camera为中心组摆放在正面视角位置，另外2个Intel RealSense和Event-based Camera分别为左右侧面组摆放在中心组的左右侧位置上，三组设备保持高度一致，水平角度分别为-30°，0°，30°。本数据集已采集动作类别为43类, 采集对象31人，涵盖5种典型场景（医院、学校、家庭、会场和养老院）下常见日常动作。共计15996段动作视频。数据集录制场地为浙江工业大学计算机视觉研究所实验室。每段视频包含：彩色图、深度图和Event的数据文件。

This dataset is primarily developed for research on and demand construction of gesture/posture action recognition and intention cognition. The dataset was recorded in the Laboratory of Computer Vision Research Institute, Zhejiang University of Technology. Three sets of equipment integrating six sensors were employed to capture human actions, comprising 3 Intel RealSense cameras and 3 Event-based Cameras. The sensor layout is designed as follows: one Intel RealSense camera and one Event-based Camera form the central group, positioned at the front-view angle; the remaining two Intel RealSense cameras and two Event-based Cameras constitute the left and right side groups respectively, placed on the left and right sides of the central group. All three groups maintain consistent height, with horizontal angles of -30°, 0°, and 30° respectively. A total of 15996 action video segments have been collected, covering 43 action categories from 31 participants, including common daily actions across 5 typical scenarios: hospital, school, home, conference venue, and nursing home. Each video segment includes three types of data files: color images, depth maps, and Event data files.

搜集汇总

背景与挑战

背景概述

手势体势数据集是一个面向手势/体势动作识别与意图认知研究的大规模多模态数据集，由浙江工业大学计算机视觉研究所创建。该数据集采用多传感器（包括RGB、深度和事件相机）采集，覆盖43类动作、31名对象和5种典型日常场景，包含15996段视频，数据总量达1.7TB，适用于计算机视觉和人工智能领域的多模态行为分析研究。