University of Twente Student Theses


Classification of damaged buildings in aerial oblique images and laser scanning data

Cyprian, Ogut Joseph (2013) Classification of damaged buildings in aerial oblique images and laser scanning data.

[img] PDF
Abstract:The purpose of this research is to classify damaged buildings on an earthquake affected site based on some of the damage classes defined in the damage catalogue. This process of damage assessment is important as it facilitates planning for disaster response organisations. Knowledge on the state of the earthquake affected site comes from study of the data captured, which is representative of the site. Input for this research was in form of two point cloud datasets; an Aerial Laser Scanning and an Aerial Oblique Photography generated point cloud. To gain an understanding of the nature of the two datasets a data analysis process was carried out. The process involved determining noise levels thus quality of the datasets. Successively, features defining unique properties of target objects of interest (i.e. walls, roofs, rubble and urban tree crowns) were determined and documented to ease in the classification process. This would help formulate the process of extracting target objects of interest. The approach of classification here was a data driven, rule based classification process. This was informed by the ability of point cloud data to capture the true form of target objects individually, and status of the building environment collectively in an earthquake affected site. Subsequently a two step approach, involving classifying target objects of interest first before performing the overall building assessment to determine the damage class or type was followed. The method was developed on the premise that individual object properties cumulatively contribute to define the final damage status of any building. The results of the classification process were then evaluated for completeness, correctness and quality to determine the success rate of the damage assessment process. This involved determination of the completeness, correctness and quality of the results using the true positive, false negative and false positive to compute the statistical values. It was found that the wall and roof classification results were relatively good, scoring 70’s and 90’s for completeness and correctness respectively. The rubble results were relatively low scoring high 60’s in completeness and correctness respectively. To conclude it all, the idea of combining two similar datasets to provide full coverage of the study area proved very successful. This presented a unique and innovative approach to capturing and analysing both vertical view or roof objects and horizontal view or facades of buildings within the dataset site. The classification process highlighted the feasible nature of combining ALS and AOP point cloud data to facilitate full site coverage, a feat extendible to any other target object of interest that is not related to the building environment.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page