Abstract

There is a growing need for rapid and accurate damage and debris assessment following natural disasters, terrorist attacks, and other crisis situations. This research enhances existing algorithms for LiDAR point classification (ground/non-ground), feature classification (buildings, vegetation, roads, etc.), and seeks to develop new algorithms for building damage and debris detection and quantification-work evaluated using LiDAR data of Port-au-Prince, Haiti, collected by RIT just days after the January 12, 2010 earthquake. Normalized height, height variation, intensity, and multiple return information are among the parameters being used to develop rules for building extraction and vegetation removal. Various approaches are being explored to perform damage assessment, with a focus on the slope and texture of roof planes. Initial results show a general over-segmentation in a 457x449 m region-of-interest (ROI)-the building detection algorithm autonomously identified 206 buildings, while only 98 buildings actually exist in the ROI. Further, four buildings went completely undetected. The accuracy of the damage detection algorithm was assessed only in regions where the building detection algorithm results overlapped actual building locations. The overall damage detection accuracy was 73.40%, but with a low Kappa accuracy of k = 0.275. The algorithms will be implemented in a common programming language where the processing will be optimized for large data sets. The goal is for the operational tool to be implemented in the field, using available equipment in a close to real-time environment