Phenomenology-based segmentation of InSAR data for building detection

Abstract

By inter ferometric SAR measurements digital elevation models(DEM) of large areas can be acquired in a short time. Due to the sensitivity of the inter ferometric phase to noise, the accuracy of the DEM depends on the signal to noise ratio (SNR). Usually the disturbed elevation data are restored employing statistical modeling of sensor and scene. But in undulated terrain lay over and shadowing phenomena occur. Furthermore, especially in urban areas, additional effects have to be considered caused by multi-bounce signals and the presence of dominant scatterers. Unfortunately, these phenomena cannot be described in a mathematically closed form. On the other hand it is possible to exploit them in model-based image analysis approaches. In this paper we propose a method for the segmentation and reconstruction of buildings in InSAR data, considering the typical appearance of buildings in the data.