In this research, an approach of three-phase 3D corresponding algorithm based on hybrid feature of Harris and line feature points matching is proposed, in which relative matching, relaxation matching and LMedS matching are implemented respectively. At the mean time there Human-machine interaction processes are added, namely limitation of matching area, adding feature points by hand as well as line matching. Through coordinated transformation between views, 3D data of scene form different views integrated into a uniform reference coordinate frame, which can solve the false matching problem caused by inconsistent or shadowed areas. Experiment demonstrates the efficiency and robustness of this corresponding method, in which the 3D model of the targeted object can be displayed on all views.