Orientation and DEM extraction from ALOS-PRISM images using the SRTM DEM as ground control
José A. Gonçalves(1)
(1) University of Porto, , , Portugal
The PRISM sensor onboard of the ALOS satellite is a three-line scanner with a potential to derive elevation data with a vertical accuracy compatible with medium scale topographic mapping scales, such as 1:25,000 or even larger. Being dedicated to systematic stereo coverage it can contribute for topographic mapping purposes in many undeveloped regions of the world. However, the need of ground control points (GCP) for the purpose of image orientation is a limitation in this process, since their acquisition, with appropriate planimetric and vertical accuracy will normally require field survey. This may be very expensive or impossible in many areas.
This paper presents a methodology of image orientation and DEM extraction without the use of GCPs. It uses the SRTM DEM, which is available for most of the Earth, as a ground control surface. A relative DEM is extracted from the images, using only the approximate image location provided. This DEM is then brought, together with the images, to its correctly georeferenced position, by matching the SRTM DEM. Although the SRTM-DEM has a relatively coarse resolution (90 m), the DEM matching can be done with a very good accuracy, of around 5 meters.
The method makes use of images in processing level 1B2-R. In this mode the original images were projected on the reference ellipsoid, using the approximate exterior orientation, an projected in a UTM map projection. Since rigorous photogrammetric equations can not be applied in this case an approximate sensor model was considered. First, UTM easting and northing coordinates are obtained from image coordinates by an affine conformal transformation. These coordinates are not correct due to relief effect, but a good estimation of this relief displacement can be obtained in the case of the nadir image. Due to the uncertainty in initial sensor orientation a positional error remains. In the absence of standard GCPs this displacement will be corrected using the control surface.
After registering the forward and backward images to the nadir image space, parallaxes in y direction for the backwards and forwards images can be obtained, allowing for height determination. Small DEM patches (4 by 4 km2) are extracted by correlation. These DEM patches are then matched to the SRTM DEM, allowing for the determination of the local displacement (around 50 meters) required to bring the extracted DEM to its correct position. This is done in several small patches, since displacements are not constant. The displacements can be modeled as function of the patch position in the image by an affine transformation. These displacements allow for the correction of the approximate sensor model.
Tests were done with a PRISM image of a hilly region in Portugal, with low vegetation, and heights ranging from sea level to 1400 meters. In order to do an independent assessment coordinates and heights of total of 56 check points were obtained from orthophotos (0.5 m resolution) a DEM, derived photogrammetrically for mapping scale 1:10,000. These points were found to have a root mean square error smaller than 5 meters in planimetry and 3.2 meters in height. The overall accuracy is compatible with the standards for 3D data extraction for 1:25,000 scale mapping. The orientation could be obtained with accuracy nearly as good as with standard GCPs.