A Tribute to Opportunity and her Epic Journey, Flying Over Endeavour Crater Options

[Nirgal]

Although this is still a work in progress, I would like to share with you a preview of a new project I'm working on:

Flying Over Endeavour Crater - A Tribute to Opportunity and her Epic Journey

(The video is best viewed in HD resolution (720p), fullscreen and volume turned up for the background music.

Click here for a (still experimental) excerpt as anaglyph (view in HD if possible)

Here is some background info for the technically interested:

The whole DEM consists of about 3 billions of triangles, modelling Endeavour's west rim at full HiRISE resolution of 25cm/pixel. Unlike the standard method (of draping hi-res 2D imagery over a lower-res 3D DEM), the new technique provides real 3D geometry down to the highest resolution level (i.e. 0.25 m/pixel for HiRISE). This results in a more realistic visualization because each single pixel contributes genuine 3D information - as you can see for example in the low sun images where each pixel is, in principle, capable of casting its own tiny shadow.

Rendering this kind of DEMs has been a real challange, though. At 0.25 m post spacing, the models are about 16 times larger than the already huge standard HiRISE DEMs at 1 m/pixel. So together with the mosaics of several full-res CTX frames ( that I'm merging the DEMs with for context ) this results in DEMs of several billions of triangles. Unfortunately this seems to be too much to render with conventional 3D programs such as 3ds-max that are usually limited to some dozens of millions of polygons, but can't readily handle several billions. This is why I decided to write a specialized gigapixel-ready terrain visualization software with a raytracing kernel for realistic soft shadows and global illumination.

As this is a rather time consuming work, it is not finished yet (still missing color support and the movie processing chain is still too slow for producing longer flyovers ... )

Hardware is another limiting factor (currently I'm running out of RAM on 24 GB ... but the new gear (48 Gigs + 12 cores) is already on the christmas wishlist...

Nevertheless the first results look promising so I thougt I'd share some impressions with the UMSF community

[Nirgal]

Thank you very much for the nice comments on my work. glad you like it !

As there were questions about the vertical scale of the model I would like to point out there is no intentional vertical exaggeration of the geometry as you can see in this image comparing a "ground truth" photograph taken by the rover as of Sol 2679 on the left hand with a simulated view of the 3D model from approximately the same position and viewing angle (the shadowed slope in the right image is due to a lower simulated sun angle)



It is true that in my renderings I always try to emphasize & accentuate the topography for reasons of viewing aesthetics, but this is largely due to my using very low sun illumination angles just by simulating late-evening conditions with long shadows (this is just the same as landscape photograpers here on earth would do in order to accentuate the shape of dunes and mountains) In addition to that I tend to use settings of my simulated camera and darkroom that result in contrast enhancement and lens-induced geometric distortions (again the same as landscape phoptographers would do here on earth as well :-)

That being said: although the large scale topography of the model is exactly registered to official MOLA and HRSC DEMs
there remains uncertainty in the small-scale variations of the surface model as with any single-image based 3D reconstruction method namely in the presence of albedo variations. Without albedo variations (i.e. fully dust- or ice- coverd areas) I found the mean reconstruction errors in my experiments to be less than ten precent compared to ground-truth DEMs but the error can be larger in presence of large albedo variations.
Fortunately, the single-image method complements nicely with stereo-based methods that are immune to albedo variations (but provide only a fraction of the resolution and have troubles in texture-less areas).

Although maximum scientific accuracy has never been the main goal of my work (that is intended for porposes of visualization and aesthetics only - in the sense of "Martian Landscape Photography" I do plan to further improve the overall accuracy of my models in the future by incorporating more stereo-derived base DEMs. I'm especially looking forward to the official Endeavour HiRISE DTM that I think will be published in the near future and that I expect to provide a very accurate base-DEM for that area

[Guest_Oersted_*]

Thanks for your comprehensive explanations Nirgal!

In addition to that I tend to use settings of my simulated camera and darkroom that result in contrast enhancement

I think that this particular facet of your post-processing should be toned down a little bit. I love the low-sun angle and the vertiginous lens work, but I think the very punchy contrasts are probably a bit overdone.

A small "artistic" suggestion: when you come down from up high with your camera you could maybe fly through a few of those wispy high clouds we have seen in MER images. Would be a good effect to complement your lower ground-hugging fog layers and atmosphere opacity.

Can't wait to see more of your amazing work!

[Nirgal]

... I love the low-sun angle and the vertiginous lens work, but I think the very punchy contrasts are probably a bit overdone.

Yes, this is a good point. Part of the problem (apart from my not always resisting the temptation of turning the knobs too much lies in the general difficulty of preparing images to look equally good on different output devices (google "gamma-correction" ) Incidently, I just now found out that the video does look quite different (too crisp and too contrasty indeed) on monitors other than the good old "warm & soft" analog CRT display that I normally use.
Another problem is video compression: After uploading to youtube, I found that the encoding/compression caused further contrast and gamma correction to the point that many subtle details in shadowed areas got lost.

As for the cloud layers: yes this is something quite high on my TODO-list. But here, too, one should be careful not to overdo it: I once saw an animation involving artifical Mars clouds that looked very unnatural (way too "earth-like"). It will be a challenge to simulate the very faint and subtle appearance of the clouds as I imagine them to look like on Mars in reality.