OBJECT ORIENTED CHANGE DETECTION OF BUILDINGS AFTER A DISASTER

Kurt T. Rudahl, Sally E Goldin

2009 · 10 citations · 5 references

Concepts

Abstract

Following a disaster, geoinformatics can provide critical information pinpointing the location and nature of serious damage. However, using geoinformatics in this situation imposes unusual constraints: opportunistic use of whatever pre- and post-disaster data may be available without regard to quality or sensor compatibility, analysis of data by relatively untrained personnel, and limited data processing facilities. This paper describes an approach to dealing with these limitations, undertaken by our Remote Sensing Laboratory at the Computer Engineering Department of King Mongkut's University in Thonburi, Bangkok, Thailand. To assess the severity of devastation, most damage assessments focus on the destruction of man-made objects, particularly buildings. Although there is a rich literature on building extraction, existing approaches often mis-detect small-size buildings. In this paper, we present a robust rectangular building detection process that can discover both small-size buildings in residential areas and large-size buildings in industrial areas with about 75 percent accuracy. We also propose an object-based change detection approach that is flexible for disaster cases in which the predisaster and post-disaster images may not be captured with the same resolution. We evaluated our methodology using data from the tsunami of December, 2004 which devastated the Andaman coast of Southeast Asia including Thailand. According to our results, we conclude that, in conjunction with accurate inputs from the building detection, our change detection can discover real changes.

References

5