Open-Structure
收藏资源简介:
Open-Structure数据集是由慕尼黑工业大学和天津大学等机构合作创建的,旨在评估视觉里程计和SLAM方法。该数据集包含22个序列,其中16个来自真实世界,6个通过模拟生成。数据集提供了2D测量、特征对应、结构线、3D地标和共视因子图等,用于评估SLAM系统中的初始姿态估计、参数化、优化和回环检测模块。创建过程中,真实世界序列基于RGB-D图像和真实姿态生成,模拟序列则通过设计多样化的轨迹和观测增强数据集的多样性。该数据集适用于SLAM算法模块的评估,特别是在避免数据预处理模块对消融实验影响方面具有重要价值。
The Open-Structure Dataset was collaboratively developed by institutions including Technical University of Munich and Tianjin University, with the objective of evaluating visual odometry and SLAM methods. This dataset comprises 22 sequences, 16 of which are collected from real-world environments and 6 are synthetically generated. It provides 2D measurements, feature correspondences, structural lines, 3D landmarks, and co-visibility factor graphs, which are utilized to assess core modules in SLAM systems such as initial pose estimation, parameterization, optimization, and loop closure detection. During its construction, real-world sequences are generated based on RGB-D images and ground-truth poses, while synthetic sequences are crafted with diverse trajectories and observations to enhance the dataset's diversity. This dataset is suitable for evaluating SLAM algorithm modules, and holds particularly significant value in avoiding the interference of data preprocessing modules in ablation studies.




