건설현장 드론 포인트클라우드와 BIM 간의 Convex Hull 기반 자동 정합 방안

초록

This study proposes a Convex Hull-based automatic registration method between BIM and drone-based point clouds for construction status analysis. Drone-based point clouds acquired from actual construction sites often contain temporary structures and non-target objects, while interior data may be incomplete due to the limitations of exterior drone imaging. These characteristics create geometric differences between BIM and point cloud data and may reduce the applicability of conventional registration methods that rely on entire shapes or local feature correspondences. To address these limitations, this study uses the lowest floor boundary, which can be commonly identified in both datasets, as the geometric reference for registration. The proposed method sequentially performs Z-axis alignment, Convex Hull-based floor boundary extraction, XY translational alignment based on the hull centroid, and rotational alignment based on the major-axis direction of the minimum bounding rectangle. Registration accuracy was evaluated using corresponding upper slab planes extracted from BIM and drone-based point clouds. The proposed method achieved a Grid IoU of 84.0%, with a mean distance error of 151 mm, an RMSE of 299 mm, and a median error of 74 mm. In comparison, ICP produced a mean error of 259 mm and an RMSE of 436 mm, while FPFH+ICP produced a mean error of 985 mm and an RMSE of 2,194 mm. These results indicate that the proposed method can provide stable spatial alignment between BIM and drone-based point clouds even under construction-site conditions involving temporary objects and incomplete interior data.

키워드