Detection of Damaged Buildings using Edge, Texture and Segmentation: A Case study of Haiti Earthquake

Abstract

This research describes the results of a combined method that consists of several different algorithms for semi-automated damage detection of earthquake affected areas of Port-au-Prince, Haiti using high resolution satellite imagery. The objective of this study was to combine information from different methodologies to enhance damage detection in earthquake affected area. Port-au-Prince, Haiti was struck with a 7.0 magnitude earthquake in January. Due to the complex urban structure of houses and commercial buildings in Port-au-Prince, it was hard to identify damaged buildings. A rapid visualization of change in urban crisis areas is an important requirement or condition for planning and coordination. This research combined change detection results from spatial filtering, spectral change analysis and Segmentation. As a first step, change detection was done using spatial content of the image. An edge detection algorithm on the images was applied and the edges in both the pre and post event images were extracted. Secondly, the Spectral/Textural content of the image was used to extract change information from both the scenes. As a third step, segmentation was applied to extract the building footprint in the post Images. The overall accuracy of 89% was achieved when results from all the three methods were combined. This technique was then applied for validation on Awaran, Pakistan which was recently hit by an earthquake and it was concluded that combined approach is better to non- complex areas; however segmentation shows better and fast results on complex urban areas.