High-Resolution DTM Generation for the ExoMars Landing Site: Advances in Co-Registration, SPG, and Multi-View SFS DTM Processing

To optimize traditional digital terrain model (DTM) generation methods, we are currently utilizing data from Mars, focusing on Oxia Planum, the landing site of the ExoMars mission [1]. This region is particularly suitable due to its extensive multitemporal coverage, essential for landing site planning and ideal for evaluation. Specifically, it allows us to assess how increasing the number of overlapping datasets enhances the quality of co-registration, signal-to-noise ratio (SNR) improvement, and densification of the 3D point cloud resulting from stereo matching.

Before ExoMars’ suspension, high-resolution DTMs of Oxia Planum had already been generated using data from current Mars-orbiting camera systems [2,3]. These SPG-derived DTMs [4] used HRSC [5,6] data based on MOLA. Subsequently, machine learning-based height estimation techniques were employed to produce DTMs of varying resolutions using HRSC, Context Camera (CTX) [8], Colour and Stereo Surface Imaging System (CaSSIS) [9], and High Resolution Imaging Science Experiment (HiRISE) [10] data [2,3]. Further refinement was achieved through super-resolution techniques, yielding even higher-resolution DTMs of the region [3].

Five years later, additional CTX and HiRISE datasets of the region have been acquired, motivating us to generate a new SPG-based CTX DTM, which will later be refined using multi-view Shape-From-Shading (SFS) techniques. We are confident that the significantly increased multitemporal coverage, particularly with CTX data, will enable improved co-registration and bundle adjustment quality. Consequently, this will not only enhance the overall robustness of the DTMs but also reduce height offsets in overlapping SPG DTMs, mitigate artefacts, increase the density of the triangulation-based 3D point cloud, and improve the SNR.

We have co-registered almost 100 of the approximately 300 CTX swaths covering the broader landing site area. Following our previous work on DTM generation for Ceres [11], we manually register each individual image, or, in the case of CTX data, each swath. For this purpose, we developed a custom tool for the Integrated Software for Imagers and Spectrometers (ISIS) that automatically generates equidistant ground control point (GCP) networks based on reconstructed SPICE information (CK (Camera Kernel) and SPK (Spacecraft and Planet Kernel)) data. These GCPs are then manually registered to the HRSC Level 5 orthomosaic using the ISIS qtie function. The process is iterative: in the first step, all nadir and slightly off-nadir CTX swaths are co-registered. The orthomosaic derived from these bundle-adjusted CTX swaths then serves as a new baseline to refine the registration of individual GCPs in the second step. Currently, we are in the third iteration, now incorporating off-nadir data.

Using the ISIS integrated function jigsaw [12], we achieve bundle adjustment results for nadir and slightly off-nadir data with sigma0 values of 0.25 or better, reducing latitude and longitude offsets of map-projected swaths to sub-half-pixel accuracy. During the bundle-adjustment, we solve for CK and SPK angles, their angular velocities, and accelerations. As expected, data acquired at higher emission angles currently exhibit slightly higher sigma0 values (0.3–0.6) due to the increased influence of the lower resolution of the underlying DTM. However, we anticipate that these sigma0 values will converge toward those of the nadir data once we use our newly generated CTX DTM for SPICE initialization in the final bundle adjustment.

While the manual registration process is time-consuming, the results are transparent, reproducible, and easily usable by other ISIS users who wish to generate their own DTMs or perform bundle adjustments on CTX data. We plan to publish the corresponding GCP network files alongside the final data products.

An additional benefit of reducing lat/lon offsets through precise registration is the enhanced SNR for orthomosaic generation or even super-resolution reconstruction. The resulting higher-resolution CTX orthomosaic is expected to facilitate the co-registration of HiRISE data with resolutions of approximately 0.25 m/px.

Preliminary visual results of the co-registration quality are shown in Figures 1 and 2. For the brightness-calibrated and controlled orthomosaic, we used 50 CTX nadir swaths. The camera shading effect or frown effect [13] was corrected using our own algorithms, and the average brightness of individual swaths was calibrated to a value of 0.1. In Figure 2, we present detailed views of a 2.28 km crater, demonstrating that the mosaic, generated from multiple overlapping CTX swaths, exhibits no misregistration-induced ghosting effects. Moreover, even before applying super-resolution techniques, we have already significantly improved the SNR due to the highly accurate registration and bundle adjustments.

Once co-registration of all available CTX data is complete, we will proceed to compute the SPG-based DTM and subsequently the multi-view SFS DTM in the coming months. For this, we will use the Ames Stereo Pipeline (ASP) [14,15], with which we have previously achieved excellent results in generating high-resolution DTMs for Ceres [11]. Both ISIS and ASP are open-source tools, ensuring that our work remains transparent and reproducible

 

References

[1] Quantin-Nataf et al. 2021. Astrobiology 21(3), [2] Tao et al. 2021a. Remote Sensing 13(16), [3] Tao et al. 2021b. Remote Sensig 13(11), [4] Gwinner et al. 2016. Planetary and Space Science 126, [5] Neukum and Jaumann 2004, [6] Jaumann et al. 2007. Planetary and Space Science 55(7-8), [7] Smith et al. 1999. Science 284, [8] Malin et al. 2007. Journal of Geophysical Research: Planets 122(E5), [9] Thomas et al. 2017. Space Science Reviews 212, [10] McEwan et al. 2007. Journal of Geophysical Research 112(E5), [11] Neesemann et al. 2025. Remote Sensing 17(3), [12] Edmundson et al. 2012. ISPRS XXII, [13] Walter et al. 2024. Earth and Space Science 11(2), [14] Beyer et al. 2018. Earth and Space Science 5(9), [15] Alexandrov and Beyer 2018. Earth and Space Science 5(10).

Figure 1: Bundle-adjusted, brightness-calibrated CTX orthomosaic of Oxia Planum using 50 co-registered nadir/off-nadir swaths (sigma0 ≤ 0.25). Residual lat/lon offsets are sub-half-pixel. Figure 2 provides zoomed-in quality assessment.

 

Figure 2: Registration quality and SNR increase demonstrated on a 2.28 km crater (17.75°N/23.94°W) in Oxia Planum. a,b,e,f,h,j: Bundle-adjusted orthomosaic (26 swaths) showing improved SNR without ghosting. c,d,g,h: Original CTX image (U06_073375_1981_XN_18N024W).