Cluster-Based Roof Covering Damage Detection in Ground-Based Lidar Data

Abstract

Abstract Efficient assessment of building damage states in the aftermath of extreme events is critical for loss estimation and forensic investigations. Recent developments in ground-based laser scanning technology allows for robust acquisition of 3D data from damaged areas; however automated techniques are needed to reduce manual data processing work and extract meaningful damage information from the point cloud data. This research tested a clustering-based method to automatically detect wind-induced roof covering damage in scans of damaged buildings. Experiments were conducted in controlled laboratory conditions to determine the best algorithm settings and also objectively evaluate the performance of the algorithm under varying conditions. Among clustering features tested, LiDAR intensity “I” resulted in the highest damage detection accuracy with the average false detection less than 5%. Combining the k-means algorithm with clustering criteria such as the “elbow” method led to automated clustering in 82% of the tests. However, in order to achieve a fully automated method, a clustering algorithm that does not require a predetermined number of clusters must be employed.