Abstract: Flatness detection of large rockfill dams is critical to structural safety.Conventional flatness detection methods suffer from low efficiency and insufficient accuracy.To address these limitations,a multi-layer random sample consensus-based detection method is proposed for the multi-level Nuozhadu large rockfill dam.The proposed method consists of three key steps: ①point cloud registration and approximate voxel filtering for data downsampling; ②a PCA-based adaptive clustering denoising approach that reduces the point cloud to two dimensions to remove noise,combined with height stratification and angular clustering to segment the main dam surface point cloud; ③optimized ML-RANSAC plane fitting with normal vector constraints and distance thresholds to achieve accurate segmentation and flatness analysis of multi-layer dam surfaces.Experimental results show that,compared with the traditional RANSAC algorithm,the proposed method reduces the plane-fitting mean square error by 43.4%and improves detection efficiency by 55.2%.It also effectively identifies local surface deviations and generates visualized deviation heat maps.By integrating stratified sampling and local model fusion,the method significantly enhances large-scale point cloud processing efficiency and detection accuracy for complex geometries,providing a high-precision,low-risk automated solution for rockfill dam quality assessment.

Key words: multi-layer RANSAC, 3D laser scanning, PCAAC point cloud denoising, rockfill dam flatness detection, point cloud processing