Human pose dataset (sit & stand pose classes)

This dataset is the output of images processed by OpenPose. OpenPose is a pose estimation software that identifies the location of key body parts and joints from images or video frames and provides a high-level understanding of the human pose through a key point delimitated outline of the body that implies the underlying pose. The dataset consists of multiple entries which report on the key points of various pose images collected from the public domain - a small sample set images is provided that depicts the OpenPose mapping interpreability of the data. The dataset consists of <strong>50,727 entries</strong>: <strong>37,748 entries </strong>are associated with a 'stand' pose classification <strong>12,979 entries </strong>are associated with a 'sit' pose classification These key points are reported in csv format (stored in an Excel book) with the following data attributes: Original pose image file name List of 18 normalised XY coordinates for each body part/joint, and the associated confidence score of its localisation as reported by OpenPose. The pose class (either <em>sit </em>or <em>stand</em>) This datset served in the preliminary experiments as part of the Master's dissertation "<em>Novel data augmentation schemes for pose classification using a convolutional neural network</em>" to faciliate a binary pose classification problem. Author: JS du Toit

本数据集为经OpenPose（姿态估计软件）处理后的图像输出产物。OpenPose可从图像或视频帧中识别人体关键部位与关节的空间位置，并通过勾勒身体关键点轮廓实现对人体姿态的高层语义解析，该轮廓可隐含底层姿态结构信息。本数据集包含多条条目，记录了从公开领域采集的各类姿态图像的关键点数据；同时附带少量样本图像，用于展示本数据集的OpenPose映射可解释性。本数据集总计包含50727条条目：其中37748条对应「站立」姿态分类，12979条对应「坐姿」姿态分类。上述关键点以CSV格式存储（封装于Excel工作簿内），包含以下数据属性：原始姿态图像文件名、各身体部位/关节的18个归一化XY坐标列表、OpenPose输出的对应关键点定位置信度分数，以及姿态类别（分为「坐姿」或「站立」两类）。本数据集曾作为硕士学位论文《基于卷积神经网络（Convolutional Neural Network，CNN）的姿态分类新型数据增强方案》预实验的组成部分，用于支撑二分类姿态分类任务的开展。作者：JS du Toit

搜集汇总

背景与挑战

背景概述

该数据集是通过OpenPose处理生成的包含50,727条人体姿态关键点数据的数据集，主要用于坐姿和站姿的二元分类问题。数据以CSV格式存储，包含图像文件名、18个归一化坐标及其置信度分数，以及对应的姿态类别。