融合图像哈希特征与运动轨迹约束的塔吊多视角图像智能筛选与三维重建方法  

Intelligent multi-view image filtering and 3D reconstruction method of tower crane based on image hash features and motion trajectory constraints

作　　者：卢昱杰 张修龙 仲涛 LU Yujie;ZHANG Xiulong;ZHONG Tao(College of Civil Engineering,Tongji University,Shanghai 200092,China;Key Laboratory of PerformanceEvolution and Control for Engineering Structures of the Ministry of Education,Tongji University,Shanghai 200092,China;Shanghai Research Institute of Intelligent Science and Technology,Tongji University,Shanghai 200092,China)

机构地区：[1]同济大学土木工程学院,上海200092 [2]同济大学工程结构性能演化与控制教育部重点实验室,上海200092 [3]同济大学上海智能科学与技术研究院,上海200092

出　　处：《建筑结构学报》2025年第4期188-197,共10页Journal of Building Structures

基　　金：国家重点研发计划(2022YFC3801700);国家自然科学基金项目(52078374);上海市Ⅰ类高峰学科资助(2022-3-ZD-04)。

摘　　要：三维实景数字模型是施工现场精细化管理的数据底座,其中以塔吊摄像头为硬件基础的图像三维重建方法兼具高效与低成本优势。为提升塔吊摄像头图像采集的质量与效率,提出了融合图像哈希特征与运动轨迹约束的塔吊多视角图像智能筛选与三维重建方法。该方法基于哈希相似度以筛除冗余图像,同时通过构建特征运动轨迹和运动模型以确保图像集的完整性,最后基于COLMAP算法重建三维模型,并定量评估图像集与点云模型质量。通过某住宅项目的算例分析表明:筛选图像集满足合理重叠度与完整度需求(平均重叠度81.44%,完整度100%),此基础上重建的三维模型完整度高、几何连续性强,其点云表面密度31.83点/m^(2),孔洞占比6.94%,整体建模效率提升约90%,满足施工场景高效三维重建需求。3D real-world digital model serves as a foundational data layer for refined construction site management,enhancing efficiency and cost-effectiveness through image-based 3D reconstruction methods leveraging cranemounted cameras.To improve the quality and efficiency for image acquisition of crane-mounted camera,the intelligent multi-view image filtering and 3D reconstruction approach that integrates image hash features with motion trajectory constraints is proposed.The approach eliminates redundant images based on hash similarity,ensures completeness via feature motion trajectory and motion modeling,and reconstruct the 3D model using COLMAP.The quality of image set and point cloud is also quantitatively evaluated.A case study on a residential project demonstrates that the filtered image set satisfies overlap and completeness criteria,achieving an average overlap of 81.44%and a completeness rate of 100%.Furthermore,the resulting 3D model exhibits high completeness and geometric continuity,with a point cloud surface density of 31.83 points/m²and a void ratio of 6.94%.Notably,this method improves overall modeling efficiency by approximately 90%,meeting the requirements for efficient 3D reconstruction in construction scenarios.

关 键 词：施工现场管理 三维重建 图像集 图像哈希 特征轨迹 

分 类 号：TU17[建筑科学—建筑理论]