Weakly-Textured Cylindrical Object Reconstruction via Multi-View Tangent Dihedral Angle Constraints

Abstract. Artificial cylindrical structures such as bridge cables and traffic poles constitute critical targets in surveying and emergency response applications, necessitating accurate 3D reconstruction. Current laser scanning solutions face limitations in cost-effectiveness and operational precision, hindering widespread adoption. UAV photogrammetry, leveraging multi-view imagery, shows promise for cylindrical reconstruction but struggles with texture-deficient surfaces and slender geometries, often resulting in distorted models due to insufficient feature matching. This paper introduces a novel framework that exploits bilateral edge features and camera optical centers to construct tangent dihedral angles (TDAs) for cylindrical objects. By establishing correspondence of homologous TDAs across multi-view images, cylindrical geometric parameters are solved under multi-perspective tangency constraints, followed by polygonal mesh generation. The proposed method eliminates dependence on surface textures, achieving millimeter-level diameter estimation accuracy and superior modeling fidelity compared to conventional Multi-View Stereo (MVS) approaches. Experimental validation on bridge cables demonstrates robust reconstruction of slender cylindrical structures under challenging imaging conditions, offering practical value for infrastructure digitization and safety monitoring.