Algorithmic clustering of LiDAR point cloud data for textural damage identifications of structural elements

This study explored the potential of combining point cloud data (PCD) and data clustering algorithms for textural damage detection of commonly seen structural elements in Taiwan. Intensity and RGB (red, green, and blue color model) information acquired by ground LiDAR (light detection and ranging) were used for clustering analysis. Four data clustering algorithms, k-means (KM), fuzzy c-means (FCM), subtractive clustering (SC), and density-based spatial clustering of applications with noise (DBSCAN) were employed to detect the textural damages and to compare the corresponding efficiency and accuracy. The structural elements being studied were rusted rolling doors representing general metal materials with corrosion, walls with tile spall off representing structural elements with erosions and physical damages, and washing finish walls with water staining representing the aging and lichen covering of structural elements. Our study results suggested that both KM and FCM gave preferable clustering performance than SC and DBSCAN. They exhibited desired accuracy for the damage/anomaly identification as well as computational efficiency, suggesting that KM and FCM were more appropriate for these types of application when PCD was used. It was also concluded that intensity rather than RGB data was more appropriate for reflecting the damaged areas. Intensity data was less interfered by environmental effects such as sunlight and rainwater when used to detect the textural changes. The clustering results were also shown to be associated with the clustering number and with the nature of the textural damage type.