Comparative Evaluation of Cross-Sectional Geometric Feature Extraction Algorithms for LiDAR-Based Inclination Detection of Lattice Steel Towers

Non-contact inclination detection based on LiDAR point clouds has become an effective approach for structural condition assessment. However, due to measurement noise, lattice structural characteristics, and discontinuous point distributions, cross-sectional point clouds of lattice steel towers often contain outliers, local missing regions, and irregular boundaries, which may affect the reliability of extracted geometric features. This study establishes a comparative framework to investigate the influence of cross-sectional feature extraction algorithms on LiDAR-based inclination detection of lattice steel towers. The universality and performance of the RANSAC and Marching Square algorithms were systematically evaluated using a 110 kV overhead transmission line operating tower. Terrestrial laser scanning was employed to acquire the point cloud data. The initial registration results were subsequently further optimized through initial point cloud registration and multi-station adjustment. Four cross-sectional slicing schemes were designed, and the two algorithms were independently applied to extract cross-sectional geometric features and calculate centroid coordinates. The tower inclination was then determined by fitting the spatial distribution of centroid points. Experimental results demonstrated that both algorithms successfully extracted cross-sectional features and achieved reliable inclination detection results, with all inclination ratios satisfying the requirement specified in DL/T 741—2019 (Code of Practice for Operation of Overhead Transmission Lines). The RANSAC-based method produced inclination ratios ranging from 8.52‰ to 8.82‰, with a variation range of 0.30‰ and a mean deviation of 0.12‰. In comparison, the Marching Square-based method showed a larger variation range of 1.00‰ and a mean deviation of 0.41‰. The results indicate that RANSAC provides better robustness against point cloud noise, local data gaps, and boundary irregularities due to its inlier–outlier discrimination capability, whereas Marching Square exhibits advantages in preserving continuous contour representations when point cloud distributions are relatively complete. This study provides practical insights into the selection and optimization of cross-sectional feature extraction algorithms for LiDAR-based inclination assessment of lattice steel towers.