Refining Line Frame Construction from 3D Point Clouds: DBSCAN and Alpha Shape Approach

The application of 3D point cloud data has gained significant traction in the field of advanced building and manufacturing. This paper presents a comprehensive methodology for refining line frame construction from indoor 3D point clouds directly. The proposed methodology integrates edge detection methods and Alpha Shape computation to achieve an accurate representation of indoor structural geometry. Extensive analyses of several scenarios of indoor scene environments have demonstrated the usefulness and robustness of the methods. The study evaluates this methodology across three distinct datasets, denoted as Cases A, B, and C, each representing varying degrees of indoor room complexity. The edge identification approach uses PCA-based geometric descriptors in conjunction with DBSCAN clustering to accurately locate and segment edge points, resulting in the creation of full wireframe models. The utilization of the Rolling Ball Pivoting algorithm in the computation of the Alpha Shape facilitates the enhancement of the wireframe structure, hence enabling accurate representation. Our evaluation demonstrates exceptional adaptability and performance across scenarios, with Case C showcasing remarkable Precision (0.914) and Recall (0.941), leading to an impressive F1 score of 0.927. This research contributes to advancing indoor scene reconstruction, offering a robust methodology for precise structural representation within interior spaces.