3차원 점군에 대한 격자 기반 곡면 인식을 통한 용접선 추출

초록

This study proposes an automatic weld-line recognition method for ship and mechanical structures with curved surfaces using three-dimensional point-cloud data. Conventional CAD-based methods cannot fully reflect fabrication errors, and CAD data may not necessarily be available. Although point clouds directly capture the as-built geometry, their large volume, noise, and missing regions hinder stable surface segmentation. The proposed method partitions the point cloud into a three-dimensional grid and extracts robust plane candidates in each cell through iterative plane fitting and outlier removal. Subsequently, curved surfaces are reconstructed by merging adjacent cells based on the angular difference between normal vectors and the perpendicular distance between surfaces. Next, boundary cells are classified into outer and contact boundaries. Finally, weld lines are generated by projecting points onto intersection lines computed from paired boundary cells of adjacent surfaces, followed by calculating representative mean points and connecting them sequentially.

키워드