RGB-D stair dataset

We provide an RGB-D dataset with stair line labels and stair step surface labels for stair detection research.
The images in the training set and the validation set are from Stair dataset with depth maps [1]. Based on this dataset, we remove the RGB images taken under no light source conditions, and finally retain 2276 RGB-D image pairs as the training set and 556 RGB-D image pairs as the validation set. All the images in the training set and the validation set are padded and scaled to 512 × 512.
To implement point cloud reconstruction, we use the RealSense D435i depth camera to collect 154 RGB-D image pairs with a resolution of 640 × 480 as the test set. During the collection, we record the camera's pose and the linear transformation relationship of depth map generation for point cloud reconstruction.
The labels include stair line labels and stair step surface labels, The stair line labels have the following format:
cls x1 y1 x2 y2/n
...
Where cls represents the class of the stair line, 0 represents a convex line and 1 represents a concave line. X1 and y1 represent the coordinates of the left endpoint of the stair line, and x2 and y2 represent the coordinates of the right endpoint of the stair line. The label of an image is stored in a text file and associated by the file name.
The stair step surface labels contain the annotation of semantic segmentation for stair step surfaces. The segmentation label is stored in a gray image with a size of 512 × 512, and the classifications include stair riser surface (pixels with 128 value in the gray image), stair tread surface (pixels with 255 value in the gray image) and background (pixels with 0 value in the gray image).
For the folder structure, the train file, val file and test file store the training set, the validation set and the test set, respectively. And in train and val files, there are images file, depthes file, labels file and segmentations file, which store the RGB images, the depth images, the stair line labels and stair step surface labels, respectively. In test file, there are images file, depthes file, extrinsicses file, pcds file, plys file, which store the RGB images, the depth images, the linear transformations for depth map generation, the point clouds with pcd format and the point clouds with ply format, respectively.

[1] Wang, Chen; Pei, Zhongcai; Qiu, Shuang; Tang, Zhiyong (2023), “Stair dataset with depth maps”, Mendeley Data, V3, doi: 10.17632/p28ncjnvgk.3

本数据集面向楼梯检测研究，提供带有楼梯线标注与楼梯台阶表面标注的RGB-D数据集。
训练集与验证集的图像源自带深度图的楼梯数据集（Stair dataset with depth maps）[1]。基于该数据集，我们剔除了无光源环境下拍摄的RGB图像，最终保留2276对RGB-D图像作为训练集，556对RGB-D图像作为验证集。所有训练集与验证集图像均经过填充与缩放处理，统一调整至512×512分辨率。
为开展点云重建研究，我们使用RealSense D435i深度相机采集了154对分辨率为640×480的RGB-D图像作为测试集。采集过程中，我们记录了相机位姿以及用于深度图生成的线性变换关系，以支撑点云重建。
本数据集的标注包含楼梯线标注与楼梯台阶表面标注两类。其中楼梯线标注的格式如下：
cls x1 y1 x2 y2/n
……
其中cls表示楼梯线的类别，0代表凸棱线，1代表凹棱线；x1、y1为楼梯线左端点的坐标，x2、y2为楼梯线右端点的坐标。单张图像的标注信息存储于文本文件中，并通过文件名与对应图像建立关联。
楼梯台阶表面标注包含楼梯台阶表面的语义分割标注。分割标注以尺寸为512×512的灰度图像存储，类别分为三类：楼梯立板表面（灰度图中像素值为128）、楼梯踏板表面（灰度图中像素值为255）以及背景（灰度图中像素值为0）。
关于文件夹结构：train、val、test文件夹分别存储训练集、验证集与测试集。在train与val文件夹中，包含images、depths、labels、segmentations四个子文件夹，分别存储RGB图像、深度图像、楼梯线标注文件以及楼梯台阶表面标注文件。在test文件夹中，包含images、depths、extrinsics、pcds、plys五个子文件夹，分别存储RGB图像、深度图像、深度图生成所用的线性变换关系、pcd格式点云文件以及ply格式点云文件。
[1] Wang, Chen; Pei, Zhongcai; Qiu, Shuang; Tang, Zhiyong (2023), "Stair dataset with depth maps"（带深度图的楼梯数据集）, Mendeley Data, V3, doi: 10.17632/p28ncjnvgk.3

搜集汇总

背景与挑战

背景概述

RGB-D stair dataset是一个专门用于楼梯检测研究的RGB-D数据集，包含训练集（2276对图像）、验证集（556对图像）和测试集（154对图像），所有图像均经过尺寸调整或高分辨率采集。数据集提供两类标签：楼梯线标签（标注凸线和凹线的端点坐标）和楼梯踏步表面标签（语义分割灰度图，区分楼梯竖板、踏步和背景），并额外支持点云重建，适用于计算机视觉和机器人导航任务。