The High Resolution Stereo Camera (HRSC, [1]) is part of the orbiter payload on the Mars Express (MEX) mission of the European Space Agency (ESA), orbiting the Red Planet in a highly elliptical orbit since January 2004. For the first time in planetary exploration, a camera system has especially been designed to meet the requirements of photogrammetry and cartography for mapping the complete surface of a planet [2]. For this purpose HRSC operates as a push broom scanning instrument with 9 CCD line detectors mounted in parallel in the focal plane of the camera. Data acquisition is achieved by five panchromatic channels under different observation angles and four colour channels. At periapsis the ground resolution of the nadir channel amounts to 12.5 m, the stereo channels are typically operated at a 2x coarser resolution with the two photometry and the four colour channels at 4x or 8x coarser resolution. The data provided by HRSC are well suited for the automatic generation of Digital Terrain Models (DTMs) and other 3D data products. Such products are of vital interest to planetary sciences. As the Mars Express mission has recently been extended the prospects for a complete topographic mapping of Mars by HRSC at very high resolution are very good, indeed.

Image matching is well researched and has been documented in the literature. In general, it is agreed that in simple terrain and with adequate image acquisition geometry very good results can be achieved by totally automated approaches. Things start to be much more complicated if more complex situations are faced, such as steep terrain, height discontinuities, occlusions, poor texture, shadows, atmospheric dust, clouds, increased image noise, compression artefacts etc., some of which are commonplace in HRSC images.

Nevertheless, automatic DTM generation from HRSC images by means of image matching has reached a very high level over the years. The systematic processing chain at DLR for producing preliminary DTMs with 200 m resolution [3] runs well and stable. In addition, several groups are able to produce DTMs using different approaches, or have developed alternative modules for parts of the DTM generation process [2]. Also, a few groups have been developing shape-from-shading techniques which have reached pre-operational efficiency.

It is against this background that the desire was expressed to compare the individual approaches for deriving DTMs from HRSC images in order to assess their advantages and disadvantages. Based on carefully chosen test sites the test participants have produced DTMs which have been subsequently analysed in a quantitative and a qualitative manner. This paper reports on the results obtained in this test, more details can be found in [4].