Direct structural element recognition from scattered point cloud data (Conference Presentation)

This study attempted to build up an automated post-processing for structural element reconstruction/recognition directly from the scattered point cloud data (PCD). The target structure specimen being scanned was the three-story RC frame constructed in NCREE South Laboratory before shacking table test was performed. Algorithms used for the element reconstruction/recognition including edge signature extraction and data clustering. Edges of the target structure were extracted directly from the raw PCD and the elements were clustered by DBSCAN to ensure the geometric appropriateness. Recognition rate and dimension accuracy were compared with blue print to quantify the recognition quality, accordingly. Study results shows that the automated post-processing can achieve 100% of recognition rate with 95% of dimension accuracy, suggesting that PCD is suitable for computer vision in recognizing structure elements.