Cross-source point cloud registration technology offers the potential to harness the complementary advantages of multiple data sources by registering and integrating point clouds from diverse origins. This paper proposes a cross-source point cloud coarse registration method based on structured features in urban scenes. Firstly, we extract adjacent plane intersection lines and vertical plane boundary lines from the vertical planes of the building point cloud. Subsequently, we construct triangles based on the intersection of vertical feature lines with the ground, and use geometric constraints and semantic information for triangle matching. Finally, quick validation and fine validation are sequentially employed to determine the optimal coarse registration transformation matrix. Our experimental results demonstrate that, in comparison to point feature-based and similar point cloud coarse registration methods, the proposed method exhibits superior average accuracy, efficiency, and robustness.