An overview of MarsSI new DEM workflow and resulting Oxia Planum mosaics
Matthieu Volat,Cathy Quantin-Nataf,Amaury Dehecq
DOI: https://doi.org/10.5194/epsc2021-163
2021-07-21
Abstract:<p>The last two decades have seen a growing number of Martian orbital missions with optical cameras of a various range of resolution. The topographic context of these images is crucial to decipher Martian surface processes. Thanks to stereo-photogrammetry, digital elevation models (DEM) can be produced from couple of images taken from different view angles.&#160;</p><p>MarsSI [1] is a french Research Infrastructure that allows the users to easily and rapidly select observations, to process raw data via proposed automatic pipelines and to get back final products which can be visualized under Geographic Information Systems. MarsSI provide automated DEM generation based on the Ames Stereopipeline software (ASP) [2].</p><p><strong>DEM generation</strong></p><p>The previously used workflow, which was used until the 1st of may, 2020, of MarsSI DEM pipeline was a straightforward use of ASP to create a DEM that would be fitted against the global MOLA DEM dataset to account for orbital imprecision in elevation and referencement. We were however not satisfied with the alignment results and artefact levels in the results.</p><p><img src="https://contentmanager.copernicus.org/fileStorageProxy.php?f=gnp.cc28e437aca062377291261/sdaolpUECMynit/1202CSPE&app=m&a=0&c=9c46c58e79a26350218c9a42ecabdd7f&ct=x&pn=gnp.elif&d=1" alt=""></p><p><img src="https://contentmanager.copernicus.org/fileStorageProxy.php?f=gnp.65987047aca064477291261/sdaolpUECMynit/1202CSPE&app=m&a=0&c=e4de0df0d92e91a07a5232e7c3c82c79&ct=x&pn=gnp.elif&d=1" alt=""></p><p>To optimize the pipeline, we first enable a bundle adjustment phase to improve future correlation between left/right image. We then consider our correlation options. Recent versions of ASP allowed to use the semiglobal matching (SGM) and derived correlators. Mars often present smooth and featureless areas that yield poor results for the traditional correlator function. The SGM correlator is a relevant way to address this issue, being able to find a correlation solution with smaller window size, which also improve computational times.</p><p>Camera models and correction are provided within ASP, as well as images reprojection using an existing DEM to account for the topography. We consider using the MOLA dataset as the reference, but approximate orbital positions means that we will not match inputs to their exact position (sometimes by a few thousand meters) and introduce errors. One solution in that case is to first create a low resolution DEM with the input images we intend to use. Since at the beginning of the DEM creation, we match and align input images, this DEM fit them well despite having an approximate map-projection. Outputting this DEM at lower resolution minimize the correlation errors. We also enable missing data filling at this point by applying large maximum search distance method, to avoid loosing data from input images. This give us a result of enough quality to correct topography in the input images. We can then use these corrected images to create the high resolution DEM. During this phase, we decrease the filtering parameters compared to the first step. The result is a less smoothed DEM having sill few artefacts.</p><p><img src="https://contentmanager.copernicus.org/fileStorageProxy.php?f=gnp.9506b977aca061087291261/sdaolpUECMynit/1202CSPE&app=m&a=0&c=16dd15ae22bf49fb899fce999b789c44&ct=x&pn=gnp.elif&d=1" alt=""></p><p><img src="https://contentmanager.copernicus.org/fileStorageProxy.php?f=gnp.ed03dc87aca060287291261/sdaolpUECMynit/1202CSPE&app=m&a=0&c=a8c268a60b4496215ce08ed05af1a4a7&ct=x&pn=gnp.elif&d=1" alt=""></p><p>We also aim to improve the alignment of the produced DEM to the MOLA dataset. This step can be accomplished by ASP&#8217;s pc_align tool. One of the first issue we have is the resolution gap between MOLA and our DEM so that alignment is a challenge. To mitigate this aspect, we switch our reference from pure MOLA reference data to the blended HRSC+MOLA dataset (200m/px). We also highligth that pc_align works best when outlier measurements can be filtered. This is usually done by invalidating points that have a distance above a threshold between the dataset and the reference. Since our orbital-based georeferencing error is quite limited, by simply vertically prealign the data and using the fast global registration algorithm , the result is improved. Using fast global registration would also mean that even if the alignement fails, it would still prevent large errors to be introduced at this step.</p><p>With only minor parametrization differences, we applied the same workflow for both CTX and HiRISE imagery. Most notably, we applied stricter alignment maximum distance replacement.</p><p><strong>Oxia Planum mosaics</strong></p><p>Aside improving the overall quality of individual products, we also designed this new workflow having in mind to create mosaic of consistant, properly aligned to a common reference, datasets.</p><p>We proceed to apply this method to 32 pairs of HiRISE imagery located on the ESA ExoMars2022 landing site, 24 of which were successfully generated. Alongside the DEM, we use ASP functions to generated a corresponding reprojection of the left image with the aim to produce both a DEM and an orthorectified imagery mosaics.</p><p>However, while we find our alignment procedure to have a precision good enough for CTX scalemosaicking of products, automated HiRISE products alignements did not achieve an acceptable level. In addition to the previous workflow, we first coregistered a CTX mosaic at 10m/px against a HRSC mosaic processed and provided by the DLR[9]. This CTX mosaic was used t o align the HiRISE products whom coregistration was then refined again, manually, using ground control points against the corresponding orthorectified HRSC imagery corresponding to the DEM.</p><p><img src="https://contentmanager.copernicus.org/fileStorageProxy.php?f=gepj.57a9a2b7aca068587291261/sdaolpUECMynit/1202CSPE&app=m&a=0&c=add713c9a6872e5ebf8b0b821944a2db&ct=x&pn=gepj.elif&d=1" alt=""></p><p><img src="https://contentmanager.copernicus.org/fileStorageProxy.php?f=gepj.51962bb7aca067687291261/sdaolpUECMynit/1202CSPE&app=m&a=0&c=a3064562c3dfe37b650c3cf3514404e0&ct=x&pn=gepj.elif&d=1" alt=""></p><p>This data was used in the context of the geological mapping exercice conducted by ESA in the summer of 2020 for the Exomars2022 mission and was the primary high-resolution layer used in the MMGIS environment deployed by ESA.</p><p><strong>Conclusion</strong></p><p>This work allows the MarsSI infrastructure to produce DEM of greater quality that will also fit better in GIS environments, while requiring no user input. We now can also consider the automated production of mosaics using CTX data, and with user input for HiRISE data. While more time-consuming, this allowed us to provide a useful hi-resolution dataset for the mapping exercise of the Exomars2022 project.</p><p><strong>Data DOIS</strong></p><p>10.48326/IDOC.PSUP.MARSSI.CTX.OXIA</p><p>10.48326/IDOC.PSUP.MARSSI.HIRISE.OXIA</p><p><strong>References</strong></p><div> <div> <div> <ul><li> <p>C. Quantin-Nataf, L. Lozac&#8217;h, P. Thollot, D. Loizeau, B. Bultel, J. Fernando, P. Allemand, F. Dubuf- fet, F. Poulet, A. Ody, H. Clenet, C. Leyrat, S. Harrisson, MarsSI: Martian surface data process- ing information system, Planetary and Space Sci- ence 150 (2018) 157 &#8211; 170, ISSN 0032-0633, doi: https://doi.org/10.1016/j.pss.2017.09.014.</p> </li> </ul><ul><li>R. A. Beyer, O. Alexandrov, S. McMichael, The Ames Stereo Pipeline: NASA&#8217;s open source software for deriv- ing and processing terrain data, Earth and Space Science 5 (9) (2018) 537&#8211;548.</li> </ul></div> </div> </div>