故宫钟粹宫区域、太和殿区域三维重建模型数据集

本数据集面向北京故宫室外建筑场景，采用基于SfM的三维重建流程采集与生成，拍摄时间为白天自然光条件下的多次实地外业，保证屋檐、墙体、地面等关键结构在不同视角和不同尺度上具有充分重叠。数据原始部分由手持相机采集的多视角彩色图像构成，并同步记录拍摄时间戳等信息，时间精度可达秒级；空间范围覆盖典型院落及周边建筑立面，结合已知尺度或辅助测量进行尺度恢复，重建结果在主要结构处具有厘米到分米级的相对空间精度。计算流程包括图像去噪及曝光统一、特征提取与匹配、增量式SfM求解相机位姿、全局束束平差优化以及稀疏点云生成等环节，并对重投影误差、特征跟踪长度、点云完整性等指标进行质量控制，剔除严重模糊、过曝、遮挡过多的图像及不稳定3D点，保证模型结构连续性与几何一致性。本数据集提供原始图像、相机外参、内参信息以及重建点云等多源数据，可为城市级文化遗产三维建模、视觉定位与导航、场景理解、图像检索以及相关深度学习方法的训练与评测提供高质量基准，对推动复杂历史建筑环境下的三维重建与智能感知研究具有重要参考价值。

This dataset focuses on the outdoor architectural scenes of the Palace Museum in Beijing, and is collected and generated using a Structure from Motion (SfM)-based 3D reconstruction pipeline. The data acquisition was carried out through multiple field surveys under natural daylight during daytime, ensuring sufficient overlap of key structures such as eaves, walls, and ground across different viewpoints and scales. The raw portion of the dataset comprises multi-view color images captured by a handheld camera, with synchronized recording of shooting timestamps and other metadata, achieving a temporal accuracy up to the second level. The spatial coverage encompasses typical courtyards and surrounding building facades. Scale recovery is performed using known scale references or auxiliary measurements, and the reconstruction results achieve a relative spatial accuracy of centimeter to decimeter level for major structural parts. The computational pipeline includes steps such as image denoising and exposure normalization, feature extraction and matching, incremental SfM for solving camera poses, global bundle adjustment optimization, and sparse point cloud generation. Quality control is conducted based on metrics including reprojection error, feature track length, and point cloud integrity, to eliminate severely blurred, overexposed, excessively occluded images and unstable 3D points, thereby ensuring the structural continuity and geometric consistency of the reconstructed model. This dataset provides multi-source data including raw images, camera extrinsic and intrinsic parameters, reconstructed point clouds, and other relevant information. It can serve as a high-quality benchmark for urban cultural heritage 3D modeling, visual localization and navigation, scene understanding, image retrieval, as well as the training and evaluation of related deep learning methods. This dataset holds important reference value for promoting 3D reconstruction and intelligent perception research in complex historical architectural environments.

搜集汇总

背景与挑战

背景概述

该数据集聚焦北京故宫的钟粹宫和太和殿区域，通过基于SfM的三维重建流程采集多视角彩色图像，生成高精度三维模型，空间精度达厘米到分米级。它提供原始图像、相机参数和重建点云等多源数据，适用于文化遗产数字化、视觉定位和深度学习研究，具有重要参考价值。