Huh. Makes you wonder if Victoria would have been the destination if HiRise images had existed when they planned the trek.

It prove once again what Rui says : this is the trip that matters, not the destination...

The huge red zone around Victoria just illustrates my point. Red is not necessarily bad. Red just means slow down and use extreme caution. We cannot change directions very easily without getting into trouble when we are in a red zone. Like I said before…while some red zones are impassable, some are highways north and south. We just need to watch out for the odd dune going in the wrong direction, the occasional sand trap from a buried crater, and potential cliffs. There is little wrong about flat pavement in between large dunes.

The shortest distance from A to B is a straight line. Because of transit time, the shortest distance is not necessarily the shortest path. Yet, often the best thing to do is try to match that straight line as close as possible and within reason. Some of those red areas have near perfect highways between the dunes. You just don't want one of those dunes blocking the path you want to take.

Furthermore, we want to do some science on the way to Endeavor. Large dunes with cobbles strewn about might be good targets. I already found a large cobble about the size of the rover much closer to mini-endeavor. I’m sure there will be plenty on the way, but I’m not sure how many of the really big ones will be available on the flattest of terrain.

We have all wheels working at this point. There is concern about one failing soon. It might be easier dragging it over pavement than through a sand layer that is relatively flat. There are just lots of different aspects to this trip that need to be considered. We just don’t want to head directly into an known type of obstacle. While some dune fields are clear obstacles, others are somewhat irrelevant despite their potential size.

Here's a side by side comparison of some of the image techniques near Erebus Crater:
(I added the track from Astro0's post to all the images for comparison)

Click to view attachment


-Mike

Here are a couple of 1/4 resolution images (click for my website) with my coloured overlay.

Looks good. In fact, considering the crudeness of what I have done, I'm astonished how well it does.

Just north of Erebus - I assume Purgatory is in there somewhere.


Where we are now.


James

Absolutely spellbinding, Mike! The oarsome foursome! (I don't know which I'd prefer to steer by, but I know which I'd prefer on my bathroom floor! )
I suppose all we need is to tweak them a bit until the path we successfully followed shows as green, and the places we got stuck are red.
The suspense is killing me. Ultreya!

James, that's looking fantastic! Just a couple of remarks: could we see a higher-resolution crop of an area that contains extensive bedrock (eg Erebus highway) so we can see how the FT does there? Also, there are lots of areas that are almost saturated red - perhaps the colour table could be tweaked to give more gradations of red so we can see more easily what's the worst of the worst and what's the best of the worst etc?

Meh, the method I used doesn't work for flat featureless land. The area around Victoria and any ground without cracks will look red.

James, I'm astonished, speaking as one of the 10% of the population who is red-green color deficient, that you chose shades of red and green that I can actually distinguish! What a great map!

What I find really interesting is that Purgatory does not particularly stand out. It's not like it's in the middle of a huge red area. It just highlights that we still need to be very careful even if we avoid the large concentrations of red.

By Jove I think you've got it! I like how you captured the big dune as no go (red) and still left the bed rock as drivable areas.
Well done!

Well, after messing with this for most of the evening, I don't think the shift differential method is going to work out. I tried using 0.5 wavelength and 1 full wavelength shifts. Here are the results:
Click to view attachment
Any bedrock that is less than 1 dune wavelength wide will get indicated as "scary" due to the margin shift. I thought I could get it to cancel out but to no avail. The big scary dunes AND the bedrock zones get indicated as "no go".

If anyone has ideas to improve this, be my guest.

Here's what I did:
1) Paste image
2) Copy image into new layer
3) Gaussian blur both images to 1/4 wavelength (in pixels)
4) Shift first image either 0.5 or 1 full wavelength (in pixels)
5) Take difference
6) Merge layers - gaussian blur 2 x wavelength
7) Increase contrast (from lowest signal - black; highest signal = white)
7) Colorize images (green --> yellow --> red)
8) Overlay (multiply) the hi-res BW with the colorized image.

Here's the image indicated above of Purgatory. Note that even the big dunes here are NOT indicated as scary when this method was applied.
Click to view attachment

*sigh*

This seems appropriate about now: Click to view attachment

-Mike

JamesCanvin - Just north of Erebus - I assume Purgatory is in there somewhere.

I compared James' image to Phil's route map and here's Purgatory in the highlighted circle.
It shows up as an unremarkable spot in this image. Check out the insets - one blank and the other with Oppy's path marked.
Click to view attachment

What does that say for finding these sandtraps in advance?

Also, here's James' map with the rover's route from Eagle to VC.
I've marked Purgatory (A) and Jammerbugt (B ).
Click to view attachment

Astro0

Bummer.

I was really hoping the Fourier transform method would solve all the problems: big dune detection, funky dune detection, finding safe bedrock zones.

Would it be possible to do some sort of double Fourier analysis? To look for a change in the regular pattern from the parallel lines? The really big sandtraps seem to have off-parallel curves. And large sweeping, well...sand-traps.

-Mike

If anyone wants to relocate the highest resolution of the area, the regional location is available here at the tip of the big red arrow in an image from two Novembers ago when the HiRISE image was first released. And then Emily's full-res cropped segment of Vostok-Erebus highway that contains Purgatory is hosted here.

FWIW

It may be self-evident, but big ripples have big troughs between them. The former present formidable (RED) barriers when they lie transverse to the desired path, but the latter provide splendid highways (GREEN) when parallel to it. So I suspect that the software had better take into account the divergence between the prevailing big ripple axis and desired axis of travel, if we are to avoid confusing color schemes.

Purgatory was a different type of ripple. The ridge of the ripple curves while the other ripples ridge is straight. If you look S, SE and E of VC there are many of these curved ripples. The reason why this algorithm does not find them is because it assumes a straight ripple. Since these are probably the major threats to the rover, very likely one would need to convolve the image with a reference to detect them. Fortunately these purgatory-style ripples are very easily seen in the HiRISE image and it will be very easy to correlate the position of the rover as seen from the ground with the orbital imagery.

I can't believe the amount of work you guys have put into this! I can't wait to see the next HiRISE image.

Paolo

The message I am getting from all this amazing analysis is that the worst is behind Oppy, and that no seriously impassable barriers present themselves going forward. The overall driveability of the surface should not change ahead. Imaging taken en route and rover self-protection software should be sufficient to deal with any tricky ripples that will be encountered along the way. At worst, perhaps a couple of days delay to disentangle and shift around.

without

Wow, Ant103! I didn't expect to see that. That is amazing! In some ways it reminds me of the effects of a recent hurricane near here. That is an image to highlight.

Yep. I was thinking about that last night. There is geometry here that is not simple to capture and analyze.

I am blown away by the recent maps. There is some very good work going on. Keep it up, people.

I'll try and make those 1/4 resolution plots for the whole of image this evening. Needed an early night last night so I just did a couple to show that it works.



Yes, I had no real idea how big the numbers would get as it worked across the image so a lot of those in the etched terrain ended up saturated red. I'll rerun it with different scaling before making the maps tonight.



Well at this stage that was never going to happen - my analysis explicitly looks for repeating E-W patterns at 8 - 28 pixel wavelengths - then plots the 'power' at that wavelength.

I think it's a reasonable assumption that big/funky dunes are more likely in 'big ripple areas'. In those areas it's also harder to navigate around obstacles and to navigate through in general. More elaborate schemes are probably possible but I'm not sure it's needed.

Purgatory was one isolated bump, not only that but it is surrounded by ripples that are small compared to some of the monsters found elsewhere, so the orange classification is what I would have expected (orange -> use caution, no blind drives like at Purgatory!). As I said before, this method with give you the areas that are generally more favourable for driving - but it doesn't mean Paolo can take his eyes off the road!

Any map that had the resolution to have every 'Purgatory' and dead end marked would end up being on a similar scale to the HiRISE image, so you might as well just use it as is!

My map should help to decide the general direction to minimise the chance of tripping over something nasty - then the HiRISE images themselves can be used to further refine the route - then the previous sols nav/pancams can be used to plot the actual drive. Simple!


James

Is the way forward now for Paolo and his team to begin choosing say 4 or 5 possible routes to take in things of scientific interest using the map(s) produced so far - sticking to Green / Yellow areas where possible.

Once these routes are selected - we can then zoom into the image and overlay some lines to show actual paths between the dunes (particlularly in any RED areas that need crossong) and taking advantage of the North-South trenches for easy driving - I suspect that way you will end up with a few very detailed routes for the MER guys to choose from.

It's the sort of 80 - 20 rule - I think you guys are 80% there - why not spend 20% doing some in-depth visual work on some routes selected from the (Fourier ?) mapping done so far

Just a thought - I'm no academic - but I'm all for taking short cuts where appropriate

Brianc

I don't think that is worth doing in too much detail very far in advance.

We could map out a great route through the maze - then they'll want to nip across a couple of ripples to get to an interesting cobble and the whole route may change.

So...does this mean that can we hit the road? Got my sick pills and the bags are already in the trunk...

You guys are just A-W-E-S-O-M-E...

I have been playing a bit with various techniques to calculate terrain, based on HiRISE images.

For those of you with Windows XP/Vista OS you can download a free software tool 'roverrouter' from http://www.navtools.nl/roverrouter.htm this tool requires the Microsoft .NET environment 3.5 SP 1 (will be automatically downloaded if you don't have it already).

Written 'quick and dirty' in VC# the tool basically is able to calculate the standard deviation of the brightness of the terrain in a complete circle around each position, size of this circle, sampling-rate, etc, can be configured from the options menu, as well as the various colors to use. After a bit of experimenting it looks like this trick gives results very much similar to the Fourier analysis mentioned earlier, however it is more easy to use on machines which do not have such software available.



Most of all, the 'rover router' can calculate a route, you can enter a track across the image, and the software will analyze this route and not only state the total distance but also the estimated surface roughness along the route (both average values and maximum values), and it can produce a graph of this.



As yet this tool is far from completed, if I can find the time I'll fiddle some more with it, trying to improve further on the techniques used, any comments are welcome offcourse, although unfortunately my spare time is always very limited.

But maybe this will just be a small help for those of us who do not have access to the big mainframes and still like to give it a try to find a route for oppy. Don't use it for operational work though, I'm not giving any guarantees

Regards,

Geert

WOW! Geert, that looks absolutely fantastic!!!!!

On y go!

-Mike

Looking at a couple images, it looks like there is a trend for funky/scary dunes seem to be orthogonal to a 300 degrees to N vector. (So the small funky-scary sections of dunes themselves seem to bend to a NE-SW parallel orientation.) The zoom in plot near Purgatory shows a couple of these, as does the image of Erebus (look in the SE portion of Erebus crater near the really big dunes). There are very small sections of a few parallel dunes that might get picked up in a Fourier analysis as a repeating wavelength. (like two or three neighboring parallel dunes orthogonal to 300 degrees - is this enough?)

So if it was possible to do another Fourier analysis using exactly the same method (same wavelengths) but at a 300 degree orientation, you might be able to pick out all the extra funky/scary parts and add those to your original EW Fourier map.

-Mike

Is it possible to identify bedrock zones with specificity? It seems to me -- without knowing anything at all about it all -- that it should be possible to use some sort of masking technique to more-or-less automatically 'trump' dune colorization with identified bedrock, if that would really be so desirable.

Just to second what James said here- I think his broad-brush analysis is just what's needed right now, rather than trying to locate every individual sand-trap. And Purgatory was not the most dangerous place we passed through, by any means, so it makes sense that it should be orange rather than red.

I had one suggested improvement, thought it might be computationally too expensive- rather than dividing the image into tiles, maybe you could calculate the power for a NxN box centered on every pixel, giving a continuous map rather than a blocky one.

Great stuff,
John.

Softness map for Etched terrain to Beagle.

Click to view attachment

The lower right corner is flat terrain, actually. It stole the reddest level to the big dunes.

This was the recipe:

Duplicate layer x 2.
Gauss 5px on the top image.
Set blend mode to Difference and combine.
Brightness and contrast to 120.
Gauss 80px.
Normalize.
Color | Map | Gradient map, using a red-yellow-green gradient.
Set blend mode to Color, save.

But I made it work for Victoria with an additional trick.

Click to view attachment

Assuming that flat dusty terrain has a middle gray tone, before the Normalize step I created a map where gray areas (flat) are bright and bright/dark areas (dunes) are dark.

Duplicate layer x 2.
Invert the top image.
Set blend mode to Difference, combine.
Gauss 80px.
Normalize.
Move the layer on top of the softness map, set blending mode to Lighten only.
Combine.

But as I said, that's tricky and still won't work for flat, dark exposed bedrock...

Hi Paolo! Thank you for allowing us to participate in even a 'small' way with the drives yet to come.

James: Great work! I don't have have the tools but was excited to see the results of your work. Thank you again for your great work - it makes rather clear at a glance areas of fast easy driving vs areas of slow more meticulous driving.

A crazy thought has been nagging at me for the past couple of days that Mike's most recent post just helped to crystallize: would it be feasible to run the same analysis but for N-S, NE-SW, E-W, and NW-SE? (Not like you have anything else to do! ) The final results could be displayed in a single image by using something akin to an Evans Plot (http://www.ncl.ucar.edu/Applications/evans.shtml). Hue could be derived by the dominate dune crest orientation [maybe with green indicating crests parallel to the drive direction and red where crests are perpendicular to the dominate driving direction of interest] and saturation being derived from the magnitude of the FA (grey / low saturation for flat areas, colorful / high saturation for large dunes).



Great thread!


-- Pertinax

(PS: Geert, that tool looks very interesting. Thank you for sharing.)

.. and the Purgatory area (about middle-left) using the shorter method. Purgatory is yellow.

Click to view attachment

That's a really good idea.

It might solve the nagging "bedrock" problem for all the image-based methods (the Fourier method shouldn't have this problem).

One could select out the bedrock, convert to a more neutral mid-dune like gray, then apply the image-method of choice.
Then re-add the bedrock, but as an even darker gray ('cause bedrock should be "safe").

Hopefully by doing something like this, nice smooth bedrock but still containing big massive dunes will get lit up as "bad". But bedrock with teensy little dunes will get lit up as "good".

As long as the bedrock faces are brighter than the dune tops, this might work. I'll play more this evening.

-Mike

I notice Purgatory was unusually bright in that HiRISE image.

Bright + green is exposed bedrock. So bright + non-green could be purgatory-like points.

And in the other direction, if there was any sort of technique that highlighted curved ripples (aka "Maze of twisty little passages, all alike") or whatever landform is both 1) deemed dangerous and 2) automatically identifiable, that, too could be overlaid on the best generalized processed image even if the specific one was bleah everywhere except where it found a match.

Wouldn't do to extend this too far, lest we get, umm, bogged down, but there ought to be two or three useful types of overlayments that could assist in the process of further if not deeper visual identifications.

I've enjoyed the discussion over the past few days. Haven't seen anything that completely trashes my route first proposed in #143 and shown crudely in #209 and #222 of this thread. Here's more detail of the first 2 km, with thanks to the usual suspect.

Hoping to superimpose the proposed path on James' fancy FT map (in #298) sometime soon.

I could be wrong, but is that not because Oppy already churned up the soil there, exposing of the underlying/fresh red soil (suspecting that the HIRISE image is from the red filter)?


-- Pertinax

E-W done, N-S easy, anything else would require a fair amount of work or a rotation of the image (OK for small areas, but v difficult for the whole image) however it would probably be better to do 2D Fourier transforms. I'm not sure what you expect to gain. The ripple orientation would be nice but apart from anomalous 'Purgatory' types (that I don't think it would pick up) they are all so close to N-S that it makes no difference.



Good idea, shouldn't be too hard to implement - it would however increase the number of Fourier transforms I do from ~2 million to ~30 Billion! So maybe not every pixel, but I suppose I could overlap the FT boxes a bit. I'm still not even sure how much smaller I could make the boxes; 64 pixels should still cover a few ripples so it might work fine. So may things to try and so little time.

I could also just use a different interpolation when blowing up the output pixels back to the full HiRISE size if people prefer the blended look. I quite like the blocks as you can see exactly what terrain was used to make that particular colour - for checking the accuracy it essential to plot it that way IMO.

James

Yup, you're producing a topo map of sorts, not a road map. And when the method is applied to the whole quadrant of possible paths to Endeavour, it's going to be really really powerful. And the road mapping definitely comes with eyeballs picking out the next few days' drive.

Shooting to get the bedrock "green" indeed is a great match for the route around Erebus... because that bright bedrock is what was easy to pick out in the orbital imagery available at the time! It's interesting that HiRise leads to shooting for entirely different "highways."

I follow you.

Really, my thinking was for the areas yet to be imaged by HIRISE and helping to quickly ID areas where the orientation of the dunes deviated from what has (and I would admit will likely remain) the dominate N-S orientation -- a not overly strong argument to the unknown and visual inspection would probably be just as useful though. I was thinking that as we eventually chase the sunrise toward Endurance, noting an area where dominate dune direction deviates from the regional norm and toward one that would at lease offer the potential for trough byways could be useful.

Again, thank you for your work!



-- Pertinax

Yeah, I suppose that would be useful. I'm tempted to wait for the new HiRISE images to come in first - I have a feeling it could all get very complicated very quickly down that route - and until I'm convinced of the benefit, I'm not sure I want to go there!

In the meantime, I've rerun things with 64x64 input pixels but also taking John Spencers suggestion and not shifting the full 64 pixels each time, just 32 (not one, I'm not ready for 10's of Billions of FT's!) I've also changed the range from green to red to stop so much of the Etched Terrain being saturated red.



Click image as usual.

The more I look at this the more optimistic I am that Oppy will make good progress.

High resolution versions coming later.

James

I was just about to email the latest raw output image to Paolo, but in the spirit of publishing raw images, it struck me that others might find it useful too. So I'll post it here:


Click for PNG (1.6Mb)

James

James's latest maps have got me wondering about where ripples are found on Meridiani and why. Are the ripples associated with Victoria? Are they accumulations of dust/sand eroded from the cliffs and blown out of Victoria (recall the origin of the dark streaks)? If that's right, why is there a smooth annulus closer to Victoria? Any thoughts on this?

It would be interesting to map out the ripples using these techniques over larger areas and look for associations with other craters.

With the new method of only shifting 32 pixels each time a small error occurred in the version I posted earlier. One line of pixels was missing at each transition from one subimage to the next (I've broken the JP2 into 15 smaller JP2's for memory reasons) I've now fixed this:


Click for PNG (1.6Mb)

This route mapping business has been really interesting. I overlaid my earlier map with the black dots on James's and they map out the same hazards and good areas - his has much more information in it but the correspondence is reassuring. His map also corresponds extremely well with experience throughout the mission. I think the basics of the problem are well understood now. Let's hope for a good HiRISE of the next bit of the journey now, to complete the job.

Phil

That's good to know Phil, thanks.

I've made up colour overlays at 1/4 of the full HiRISE resolution for the whole image so you can see how it's doing.

I'm quite frankly astonished at how good it is at picking out the ripple size. The resolution of this new version is now good enough that it can (and does) pick out the more isolated large ripples, I think even Purgatory is edging towards "red". I'm really pleased with how it's turned out, hopefully it will come in useful for the adventure ahead.



Enjoy. I could spend hours looking at them but I really must go to bed...

GO OPPY!

I'm very impressed with all the analyses so for. However, I wonder about some of the assumptions that small close together ripples are good, and big far apart ripples are bad. If I recall correctly, we drove next to some big ripples that had wide flat half pipes beside them. We certainly took advantage of bedrock with widely space dunes also near by. What we need for driving is wide flat bottomed half pipes, or bed rock, that don’t dead end in a ripple crest. I wonder if most of the analyses to date are on too large a scale—multiple ripples—many pixels. What we need is the steepness of each pixel from context clues or from stereo images. Then you color the flat parts green and the steep parts of the ripple sides red. Then the path is like the mazes in the Sunday paper—follow the green space between the red lines (or white on black) to get to the heart of the maze.

What do you think?

Click to view attachment

Face on Mars! Face on Mars!

I think you're right. If there were a way to estimate the size (amplitude) of the ripple compared to the base it would be much better.

I'm running on the assumption that big wide ripples mean big tall crests. Since the route from Victoria to Endeavor is roughly SE, we will need to make a lot of dune crossings. So even if we run south on a big wide half-pipe for a while, we will need to cross the dune crests from time to time (probably much more often than the S trek towards Victoria). I'd like to avoid any dune fields if possible and run over flat easy terrain straight to the destination (or easily detour to any science targets).

"Oppy don't surf." (apologies to Apocalypse Now)

-Mike

Impressive!

Since it's a mathematical method, it should allow matching the scale between several HiRISE images under different conditions.

BTW, I tried to match the distant peaks of Endeavour as seen from Cape Verde with an HRSC image. Is this right?

I think that if you keep challenging us, we are going to need to send our own high-res camera to Mars.

Click to view attachment

It would need tweaking, but it shows that it can be done.

What I did:

After blending the original and the slightly blurred copies with the Difference mode, and adjusting brightness and contrast, I kept a copy apart, inverted it, duplicated it, blurred one to 5px, the other to 20px, blended both together with Darken only mode and normalized them. That creates a map where both the dune tops and the exposed bedrock are dark.

Then I continued the same method as before with another two copies, but instead of applying a red-yellow-green gradient in the end, I applied a transparent to white gradient. The result is a map where small dunes and bedrock are bright and bigger dunes are transparent. I put that layer on top of the previous and blended it, so only the dune tops were left as dark.

Then applied a red-orange-yellow-green gradient and blended it with the original image.

(man, it's painful to do these things on a 750MHz)