minimal
收藏资源简介:
最小纹理RGB-D SLAM数据集(Minimal Texture Dataset for RGB-D SLAM)由德国航空航天中心(DLR)使用Intel RealSense D435i相机录制,包含低纹理概念几何场景和沙地场景,并配有Vicon运动捕捉系统提供的毫米级精度真值轨迹。该数据集最初用于评估SID-SLAM算法(IEEE RA-L 2023),现作为权威版本重新上传,并扩展了首个录制会话的全部序列及此前未公开的简单序列。数据集包含两个录制日期的33个序列:2021-10-08(15个序列,包括平行线、圆、六边形、三角形、变形模型、沙地、带圆的方形以及环路变体)和2021-10-14(18个序列,包含先前在Zenodo发布的14个序列以及额外4个序列)。每个序列提供两个文件:一个ROS1 bag文件(bz2压缩,包含彩色图像rgb8 1280x720@30Hz、对齐到彩色的深度图16UC1毫米单位、相机内参、IMU数据约200Hz、深度到彩色外参)和一个TUM格式的真值轨迹文件(时间戳、位置、四元数)。此外,仓库根目录提供Vicon标记到相机彩色光学帧的外参矩阵。颜色内参(fx=908.479248, fy=907.841919, cx=650.889160, cy=364.692657)在两次会话中一致,对齐深度流共享相同内参。该数据集适用于评估RGB-D SLAM方法在低纹理环境下的表现,尤其适合视觉里程计、同时定位与建图、机器人等任务。数据采用CC BY 4.0许可证发布。
The Minimal Texture Dataset for RGB-D SLAM is recorded by the German Aerospace Center (DLR) using an Intel RealSense D435i camera. It features low-texture geometric scenes and sand scenes, with millimeter-accurate ground truth trajectories provided by a Vicon motion capture system. Originally used to evaluate the SID-SLAM algorithm (IEEE RA-L 2023), this authoritative version has been re-uploaded with extended sequences from the first recording session and previously unreleased simple sequences. The dataset contains 33 sequences from two recording dates: 2021-10-08 (15 sequences including parallel lines, circles, hexagons, triangles, deformed models, sand, squares with circles, and loop variants) and 2021-10-14 (18 sequences, consisting of the 14 previously released on Zenodo plus 4 additional sequences). Each sequence provides two files: a ROS1 bag file (bz2 compressed, containing color images rgb8 1280x720@30Hz, depth images aligned to color 16UC1 in millimeters, camera intrinsics, IMU data at ~200Hz, and depth-to-color extrinsics) and a TUM-format ground truth trajectory file (timestamps, positions, quaternions). Additionally, the extrinsics matrix from the Vicon markers to the camera color optical frame is provided in the repository root. The color intrinsics (fx=908.479248, fy=907.841919, cx=650.889160, cy=364.692657) are consistent across both sessions, and the aligned depth stream shares the same intrinsics. This dataset is suitable for evaluating RGB-D SLAM methods in low-texture environments, particularly for visual odometry, simultaneous localization and mapping (SLAM), and robotics. The data is released under the CC BY 4.0 license.
Minimal Texture Dataset for RGB-D SLAM
数据集概述
该数据集是用于RGB-D SLAM评估的低纹理数据集,由德国航空航天中心(DLR)使用Intel RealSense D435i相机采集,包含低纹理概念几何场景和沙地场景的RGB-D序列。数据集的毫米级精度真实轨迹由Vicon运动捕捉系统提供。该数据集最初为评估SID-SLAM(IEEE RA-L, 2023)而引入,本仓库为原始Zenodo版本的权威重新上传,并扩展了完整的首日记录会话及先前未发布的简单序列。
数据集内容
序列按采集日期分组:
- 10-08/:第一会话(2021-10-08),包含15个序列(parallel_lines, circles, hexagon, triangle, deformation_model, sand, square_with_circles及其循环变体)
- 10-14/:第二会话(2021-10-14),包含18个序列(包含Zenodo上发布的14个序列及forward_backward_01, circle_only_01, lru_01, ardea_01)
每个序列包含两个文件:
.bag文件:ROS1 bag(bz2压缩),包含RealSense D435i数据流:- 彩色图像(rgb8, 1280x720 @ 30 Hz)
- 对齐深度图像(16UC1, 毫米单位, 1280x720 @ 30 Hz)
- 相机内参信息
- IMU数据(约200 Hz)
- 深度到彩色外参
_groundtruth.txt文件:Vicon真实轨迹(TUM格式,时间戳+位置+四元数),可直接用于评估,与bag时间戳一致
仓库根目录提供Vicon标记到相机光学坐标系的变换文件 T_vicon_camera.txt。
标定参数
两个会话使用相同的彩色相机内参,零畸变(plumb_bob),分辨率1280x720:
- fx = 908.479248, fy = 907.841919
- cx = 650.889160, cy = 364.692657
对齐深度流与彩色流共享相同内参。深度值除以1000转换为米。
相关资源
- 数据集主页:https://www.dlr.de/en/rm/research/publications-and-downloads/datasets/minimal-texture-dataset-for-rgb-d-slam
- 已集成至VSLAM-LAB(数据集标识为
minimal):https://github.com/VSLAM-LAB/VSLAM-LAB - 原始Zenodo版本:https://zenodo.org/records/10453700
- 引用论文:https://ieeexplore.ieee.org/document/10058038
许可协议
数据集采用CC BY 4.0许可协议。




