数据堂—116,048组3D手势数据

该数据集包含116,048组3D手势数据，每组有手部mask图（RGB，24位）、深度图（16位）、相机内参文件（TXT）、三维关键点文件（OBJ）、网格文件（OBJ）、手势标签文件（TXT）、关键点demo图（JPG）、网格demo图（JPG）。采集环境为室内、右手（无手持物），覆盖第一人称和第三人称视角，多种手势类型、手指姿态和手部整体旋转姿态，多人采集。采集设备为多台Kinect，不含个人人脸敏感信息，手部mask图和深度图已对齐。该数据集可用于手势识别、手部三维重建、手部关键点检测等任务。

This dataset contains 116,048 sets of 3D gesture data. Each set includes hand mask images (RGB, 24-bit), depth maps (16-bit), camera intrinsic parameter files (TXT format), 3D keypoint files (OBJ format), mesh files (OBJ format), gesture label files (TXT format), keypoint demo images (JPG format), and mesh demo images (JPG format). The data collection was conducted indoors, targeting right hands without handheld objects, covering first-person and third-person perspectives, diverse gesture types, finger poses and overall hand rotational poses, with multiple participants involved. Multiple Kinect devices were used for data collection. The dataset does not contain any personal facial sensitive information, and the hand mask images and depth maps have been properly aligned. This dataset can be applied to tasks such as gesture recognition, 3D hand reconstruction and hand keypoint detection.

搜集汇总

背景与挑战

背景概述

该数据集包含116,048组3D手势数据，每组数据涵盖手部掩码图像、深度图、相机内参、3D关键点、网格文件及手势标签等文件。数据采集于室内环境，使用多台Kinect设备，针对右手（未持物）采集了第一人称和第三人称视角下的多种手势、手指姿态和手部旋转姿态，适用于手势识别、3D手部重建和关键点检测等任务。