- Automatic spatial matching of multi sensor data

Context and objectives

This project aims at the interpretation of multi sensor information within monitoring and warning systems. The diversity of sensors that observe the earth all give a unique picture of a certain phenomenon, as well in ground resolution as in the spectral range of the imagery. One sensor by itself is inadequate to give a complete interpretation of the observed scene. Having data available from complementary sensors permits a more accurate and detailed analysis.

Project outcome

Expected scientific results

A prototype system has been developed based on scale-invariant shape description and matching techniques, which is able to find correspondences between vector descriptions of object contours. These descriptions can come from a GIS database (e.g. a river or coastline represented by a polyline) or can be extracted from images (e.g. a coastline represented as a chain of pixels). Given these descriptions the system finds correspondences between similar parts. These correspondences can then be used to calculate the transformation between the image or between the image and the GIS database. The system has been tested and shows robustness for shape deformations due to affine transformations, partial cover and image noise. This makes the system suited for registering images with different spatial resolution (e.g. Landsat TM and Resurs). The feasibility of automatic registration of low and high resolution and of images and GIS-data was experimentally verified. The quality of registration depends on the spatial distribution of the control objects over the images. In order to guarantee an even distribution over large areas, different types of objects need to be extracted from the image (e.g. coastlines, rivers, boundaries of areas like cities, forests and fields). In this project we concentrated on coastlines but the system easily extends to other objects.

One task within the project was reserved on finding new ways to improve existing techniques that are traditionally used within monitoring. This resulted in the following case studies : 1) refinement of medium resolution images; 2) supporting system requirements for monitoring; 3) shape analysis for the characterisation of changes.

Location: Region:
  • Grèce

Related publications: