Wireframe Extraction of Urban Linear Objects from Aerial Lidar Point Clouds

Wireframe extraction of urban linear objects from aerial lidar point clouds is important for 3D city modelling, infrastructure inspection, and asset management. Although power-line and pylon reconstruction has been widely studied, reliable type-agnostic wireframe extraction remains difficult because aerial point clouds are sparse, incomplete, and structurally complex. To address this gap, we present an exploratory comparison of four representative strategies—3D RANSAC, 3D–2D RANSAC, Region Growing, and Hough Transform—and propose a pairwise Markov Random Field (MRF) formulation optimised by graph cuts. The methods are evaluated on Dutch aerial lidar data using manually delineated reference wireframes. Results reveal clear trade-offs among the baselines. On the selected difficult pylon case, the proposed method achieves the lowest angular RMSE and qualitatively retains some internal members, but its high unmatched rate indicates that many false-positive edges remain. We identify the remaining challenges of wireframe extraction from sparse point clouds and discuss directions for more robust hybrid solutions. The implementation is publicly available at https://github.com/Ganbusier/final_thesis.