ESA       
   workshops       
envmail@esa.int

 

Georeferencing and Orthoimage Generation from Long Strips of ALOS Imagery

Clive Fraser(1), Franz Rottensteiner(1) and Thomas Weser(1)

(1) Cooperative Research Centre for Spatial Inform., University of Melbourne, 3010 Melbourne, Australia

Abstract

Imagery from the ALOS PRISM and AVNIR sensors offers the potential of georeferencing and orthoimage generation to support medium- and small-scale mapping. In order to optimize the economy and productivity of these operations, image strips of maximum possible length need to be employed with a minimum of ground control, without suffering a loss in attainable accuracy. This paper describes the development of a computational system that enables realisation of this goal. The data processing flow commences with an image merging stage that results in a single scene for AVNIR and three single scenes for PRISM. In the case of PRISM, Level 1B1 sub-images are merged, and for AVNIR individual image bands are merged. Along with the merging of imagery, all sensor orientation data is rigorously transformed and referenced to a single orbit, with one set of orbit path and attitude parameters. A strip adjustment is then carried out to refine the sensor orientation. This employs either four or more surveyed ground control points (GCPs), or image matching to underlying orthoimagery at the two end regions of the strip. The latter approach removes the need for surveyed GCPs altogether, though it requires a digital elevation model to define heights. This refined sensor orientation modelling accounts for biases within the orbit and attitude data, resulting in 1-pixel and even sub-pixel level georeferencing accuracy for all scenes of the strip.

The paper overviews the computational models and steps involved, for both PRISM and AVNIR imagery, and describes how the workflow has been accomplished within the Barista software system. Practical experimental test results are presented, in which it is demonstrated that 1-pixel level accuracy and better is achievable with as few as four GCPs over strip lengths exceeding 150km in length. Such a GCP requirement is well below that generally needed for orthoimage generation approaches that neither employ rigorous sensor orientation modelling for ALOS imagery nor have the capability to merge along-track scenes into long continuous strips. An adjustment process to refine the relative alignment and offsets between PRISM sub-scenes is also possible. At the price of some additional ground control, the refinement of relative CCD chip alignment at regular intervals over a test range with many GCPs could improve the calibration of interior orientation for each PRISM camera and further improve the accuracy of georeferencing.

 

Symposium presentation

 

  Higher level                 Last modified: 07.05.06