Object-based SAR change detection for security and surveillance applications using density based clustering

Abstract

Unsupervised change detection with high-resolution SAR images is a powerful tool for security and surveillance applications. Most change detection methods simply detect which pixels changed between a pair of SAR images, and are unable to distinguish different types of changes. In this paper we show how density based clustering can be used for grouping together the changed pixels that belong to the same object, detecting in this way which objects changed. These changed objects can then be classified into different categories by analyzing object attributes such as shape and texture and taking into account prior knowledge about the scene.