MSRA Hand

MSRA Hands 是用于手部跟踪的数据集。使用英特尔的创意交互式手势相机总共捕获了 6 个受试者的右手。每个受试者被要求在 400 帧的视频序列中做出各种快速手势。为了考虑不同的手尺寸，为每个主题指定了全局手模型比例：主题 1~6 分别为 1.1、1.0、0.9、0.95、1.1、1.0。相机内在参数为：主点=图像中心（160,120），焦距=241.42。深度图像为 320x240，每个 .bin 文件按行扫描顺序存储深度像素值，即 320240 个浮点数。单位是毫米。 bin 文件是二进制文件，需要使用 std::ios::binary 标志打开。 joint.txt 文件存储 400 帧 x 每帧 21 个手关节。每条线有 3 * 21 = 63 个浮点数，用于 (x, y, z) 坐标中的 21 个 3D 点。 21 个手关节是：手腕、index_mcp、index_pip、index_dip、index_tip、middle_mcp、middle_pip、middle_dip、middle_tip、ring_mcp、ring_pip、ring_dip、ring_tip、little_mcp、little_pip、little_dip、little_tip、thumb_mcp、thumb_pip、thumb_dip、thumb_tip。对应的 *.jpg 文件仅用于深度和地面实况关节的可视化。

MSRA Hands is a dataset for hand tracking. The right hands of 6 subjects were captured using Intel's Creative Interactive Gesture Camera. Each subject was asked to perform a variety of rapid gestures in a 400-frame video sequence. To account for varying hand sizes, global hand model scale factors are assigned to each subject: Subjects 1 to 6 correspond to 1.1, 1.0, 0.9, 0.95, 1.1, and 1.0 respectively. The camera's intrinsic parameters are: principal point = image center (160, 120), focal length = 241.42. The depth images have a resolution of 320×240. Each .bin file stores depth pixel values in row-scan order, containing a total of 320×240 floating-point numbers, with the unit being millimeters. The .bin files are binary files that must be opened with the std::ios::binary flag. The joint.txt file stores data for 21 hand joints across 400 frames. Each line contains 3×21=63 floating-point numbers, corresponding to the (x, y, z) coordinates of the 21 3D hand joints. The 21 hand joints are listed as follows: wrist, index_mcp, index_pip, index_dip, index_tip, middle_mcp, middle_pip, middle_dip, middle_tip, ring_mcp, ring_pip, ring_dip, ring_tip, little_mcp, little_pip, little_dip, little_tip, thumb_mcp, thumb_pip, thumb_dip, thumb_tip. The corresponding *.jpg files are only used for visualizing the depth maps and the ground-truth joint positions.

搜集汇总

背景与挑战

背景概述

MSRA Hand是一个用于手部跟踪的数据集，包含6个受试者的右手深度图像和3D关节标注数据，每个受试者提供400帧视频序列，涵盖各种快速手势。数据以二进制.bin文件存储320x240深度图像（单位毫米），并附带joint.txt文件记录每帧21个手关节的3D坐标，适用于计算机视觉中的姿态估计和手势估计任务。该数据集由Chinese University of Hong Kong和Microsoft Research于2014年发布，支持手部跟踪算法的研究和开发。