A Method for detection and quantification of building damage using post-disaster LiDAR data

Show full item record

Redirect: RIT Scholars content from RIT Digital Media Library has moved from http://ritdml.rit.edu/handle/1850/14462 to RIT Scholar Works http://scholarworks.rit.edu/theses/3023, please update your feeds & links!
Title: A Method for detection and quantification of building damage using post-disaster LiDAR data
Author: Labiak, Richard
Abstract: There is a growing need for rapid and accurate damage assessment following natural disasters, terrorist attacks, and other crisis situations. The use of light detection and ranging (LiDAR) data to detect and quantify building damage following a natural disaster was investigated in this research. Using LiDAR data collected by the Rochester Institute of Technology (RIT) just days after the January 12, 2010 Haiti earthquake, a set of processes was developed for extracting buildings in urban environments and assessing structural damage. Building points were separated from the rest of the point cloud using a combination of point classification techniques involving height, intensity, and multiple return information, as well as thresholding and morphological filtering operations. Damage was detected by measuring the deviation between building roof points and dominant planes found using a normal vector and height variance approach. The devised algorithms were incorporated into a Matlab graphical user interface (GUI), which guided the workflow and allowed for user interaction. The semi-autonomous tool ingests a discrete-return LiDAR point cloud of a post-disaster scene, and outputs a building damage map highlighting damaged and collapsed buildings. The entire approach was demonstrated on a set of six validation sites, carefully selected from the Haiti LiDAR data. A combined 85.6% of the truth buildings in all of the sites were detected, with a standard deviation of 15.3%. Damage classification results were evaluated against the Global Earth Observation - Catastrophe Assessment Network (GEO-CAN) and Earthquake Engineering Field Investigation Team (EEFIT) truth assessments. The combined overall classification accuracy for all six sites was 68.3%, with a standard deviation of 9.6%. Results were impacted by imperfect validation data, inclusion of non-building points, and very diverse environments, e.g., varying building types, sizes, and densities. Nevertheless, the processes exhibited significant potential for detecting buildings and assessing building-level damage.
Record URI: http://hdl.handle.net/1850/14462
Date: 2011-08-04

Files in this item

Files Size Format View
RLabiakThesis8-4-2011.pdf 18.54Mb PDF View/Open

The following license files are associated with this item:

This item appears in the following Collection(s)

Show full item record

Search RIT DML

Advanced Search