포인트클라우드 기반의 실내 공간 구조체 식별 방안

초록

Point cloud data can be used to automatically identify real indoor spaces. However, indoor environments often include many non-structuralelements. This study presents a method to remove non-structural elements from point clouds gathered through 3D scanning and to isolatestructural components. Preprocessing techniques, such as SOR and voxel downsampling, were applied to optimize the data. The RANSACalgorithm detected horizontal and vertical planes, while the DBSCAN algorithm identified columns. To assess the method's performance,quantitative analyses using RMSE and M3C2 were conducted, referencing the BIM model. Results showed high precision, with RMSE valuesof 0.860628 for type 1 and 0.322795 for type 2. Additionally, M3C2 analysis indicated that type 2 had a distribution closer to a normalcurve, suggesting more stable registration results compared to type 1. The proposed approach improves the accuracy of identifying structuralelements in 3D-scanned point cloud data. However, limitations exist in applying these results to complex or irregular structures and variedspatial conditions. Future research will explore deep learning-based classification and automatic correction algorithms to recognize irregularstructural forms, aiming to broaden the method's applicability.

키워드