Guided segmentation of large-scale indoor spaces using door point clusters

Forensic-Information/Investigation-Modeling (FIM) faces significant challenges in efficiently segmenting large-scale indoor spaces needed for detailed forensic analysis in the construction industry. Traditional segmentation methods are computationally intensive and sensitive to the selection of initial seed points, often leading to inconsistent and inaccurate results. To address these limitations, this study proposes an innovative 3-module approach to segment large-scale indoor space into smaller functional spaces utilizing door points as seed points for region-growing methods alongside the typical point cloud segmentation methods: RANSAC and DBSCAN. A case study was conducted on the sampled S3DIS (Stanford 3D Indoor Scene) dataset, comprising 15 individual indoor spaces from Area 6. The results showed that the proposed approach effectively segments large-scale indoor space into semantically meaningful individual rooms. The four performance metrics, Precision, Recall, F1-Score, and IoU derived from the segmentation results of the proposed method mostly scored around 0.900, validating the robustness of the proposed method. The guided-segmentation method contributes to the industry by facilitating effective and prompt point cloud processing necessary for detailed analysis and BIM model generation.