Modeling and Information Processing
ANAÏS 3D - Analysis in 3D imaging
3D SAR platform
The use of SAR imaging meets the needs for interpreting sites under unfavorable conditions. The radar is an active sensor and can acquire an image in all weather and by day or night. The objective in the long term is to develop a sequence of processes that produce a digital elevation model consisting of the digital terrain model and a 3D model of the objects above the ground. The automatic creation of the elevation model, difficult because of the noise present on the images, is still in the research stage.
The extraction of 3D data from SAR imaging has been studied at the DTIM for several years now, notably in E. Simonetto's thesis on [1] radargrammetry. Here, we present the results of capitalizing on those studies in a software platform based on a detailed phenomenological study of buildings, which allows us to make use of radargrammetric pairs and RAMSES interferometric pairs. Its optimum operation assumes that we have an interferometric image and a stereoscopic pair of the area of interest, with these images being geo-referenced; but the possibilities offered by the different channels taken separately are also studied.
The detection of buildings is interactive; recognition is automatic and is based on the assumption that buildings have a characteristic signature and a strong echo on the ground caused by the dihedral effect [1]. The solution selected for recognizing the "L" is a generic algorithm principle named Arocs (meaning “recognition algorithm by surface criteria optimization”) [1].
The altitude data is supplied by radargrammetry [1]. Remember that the general principle is to find the point in the right image corresponding to a point in the left image that optimizes a correlation function.
|
Left
amplitude |
Interferometric
Image |
Right
amplitude |
Visible
image |
|
Detection of the dihedral with the ground (green)
and the edge of the roof (yellow) |
profiles monoscopic(green),
interferometric(blue),
radargrammetric (red) |
The results obtained by merging the three monoscopic, interferometric and stereoscopic channels are promising and confirm the need to automate the monoscopic and stereoscopic part (limiting noise, analysis of strong echoes, phenomenology, establishing matches, etc.).
Publications
[1] E. Simonetto. Extraction 3-D de structures industrielles sur des images Ramses haute résolution par radargrammétrie. Thèse de doctorat, Université de Rennes 1, 2002.
[2] L. Fouque.Sur une méthodologie générale de fusion de Dempster-Shafer pour la classification markovienne d'images multirésolutions. Thèse de l'Université de Paris VI, (2002).
