In: Stilla U, Rottensteiner F, Paparoditis N (Eds) CMRT09. IAPRS, Vol. XXXVIII, Part 3/W4 — Paris, France, 3-4 September, 2009
A PROOF OF CONCEPT OF ITERATIVE DSM IMPROVEMENT THROUGH SAR
SCENE SIMULATION
D. Derauw
Signal and Image Centre - Royal Military Academy, Renaissance Av., 1000 Brussels, Belgium - dderauw@elec.rma.ac.be
Centre Spatial de Liège - Université de Liège, Avenue du Pré Aily, 4031 Angleur, Belgium - dderauw@ulg.ac.be
KEY WORDS: SAR, Scene, Simulation, DEM, reconstruction
ABSTRACT:
In Very High Resolution (VHR) Synthesis Aperture Radar (SAR) context, very fine and accurate georeferencing and geoprojection
processes are required. Both operations are only applicable if accurate local heights are known. 3D information may be derived from
SAR interferometry (InSAR), But in VHR context, InSAR reveals to be inaccurate mostly due to phase unwrapping problems and to
phase/height noise. Generated InSAR Digital Surface Models (DSM) can only be considered as a first good approximation of the
observed surface. Therefore, we proposed to start from the InSAR DSM, to project it on ground range on a given datum, to model the
observed scene using this projected DSM, then to simulate in slant range the intensity image issued from this structure model.
Comparison between simulated and observed intensity image can then be used as a criterion to modify and improve the considered
underlying DSM.
In this paper, we present the different steps of the proposed approach and results obtained so far, showing that the proposed process
can be run iteratively to modify the DSM and reach a stable solution.
1. INTRODUCTION
A cooperation programme named ORFEO (Optic and Radar
Federated Earth Observation) was set up between France and
Italy to develop an Earth observation dual system, optic and
radar, with metric resolution. Italy is in charge of the radar
component (COSMO-Skymed), and France of the optic
component (PLEIADES).
Beside ORFEO, an accompanying programme was set-up to
prepare the use and joint exploitation of images that will be
provided from this satellites constellation. In the frame of this
accompanying programme, the Belgian Science Policy
(BelSPo) is financing the EMSOR project aiming at performing
man-made object detection for urban map updating using VHR
SAR and optical data.
While such objective is well addressed in the optical imagery,
this topic stays highly challenging in SAR imagery due to
inherent peculiarities of SAR acquisition and imaging mode.
Main obstacles are geometrical on one side and linked to SAR
signal content on the other side. Geometrical deformation
specific to SAR systems, i.e. layover, foreshortening,
shadowing, make man-made structures appearing very
differently in shape with respect to their appearance in optical
imagery (Balz T. 2003).
Specificities of SAR signal, mainly speckle, radar cross section
dependence with incidence angle and multiple reflection
processes make identical objects appear sufficiently differently
to compromise, or make inoperative, classical detection
techniques applicable in optical imagery. Man-made structures
detection in SAR images based on speckle filtering followed by
image segmentation is not applicable as such. Classification is
often considered as a first processing step that, combined with
other information layers, is used in higher level processing for
fine Digital Surface Model (DSM) extraction and man-made
structure detection (Tison et al. 2007, Thiele et al. 2007). SAR
scene simulation was also proposed to help in fine
georeferencing process (Blaz T. 2006) or to iteratively steer
building structures detection and identification (Soerger et al.,
2003).
Similarly, in this paper, we propose an iterative way to improve
a seed DSM that is obtained through classical Interferometric
processing of single pass VHR SAR data. We developed a basic
SAR intensity image simulator adapted to very high resolution.
This one is then used to improve our seed DSM, comparing the
simulated image in intensity with the detected one and using
this comparison to perform blind DSM corrections without any
a priori knowledge of the underlying urban structure.
The proposed approach is justified by the fact that classical
interferometric SAR (InSAR) is showing its limits in the VHR
context. Therefore, on-ground projected InSAR DSM can be
considered as a first approximation of the 3D observed surface
and be used as a seed DSM to be improved.
The main aim being man-made structure detection,
improvement means here reaching a DSM representation
allowing better detection and localisation of searched structures.
This paper describes first results obtained and choices that have
been made up to now to assess the validity of the proposed
iterative process. Our first aim was to perform a proof of
concept of the proposed approach, i.e. DSM improvement based
on iterative comparison between a simulated and detected SAR
intensity image.
2. TEST SITE AND SEED DSM
2.1 Data set description
To generate our seed DSM, we are using a VHR InSAR pair
acquired in February 2006 above Toulouse (France) by the
RAMSES X-band sensor (Dupuis et al. 2000). Resolution cell
dimensions are 0.55m in azimuth by 0.35m in slant range. We
