Be careful with distances taken from HiRISE images, as a lot depends on the projection which was used (if the image is map aligned), basically you can't just take pixel-distances and multiply it with the scale. Scales might vary with the latitude of the image, depending on the projection type, and directions might also not be continuous across the image. Every projection-method has its own peculiarities.

In fact it is nice that oppy landed quite close to the equator, which make chart projections a bit easier and the resulting distance-errors smaller, at the Phoenix latitude you would have far more problems calculating distances. This is one of the reasons why I try to relate everything to latitude/longitude coordinates only.

It's quite true that scale and direction will vary across a map depending on the projection used, but for the minute area covered by one HiRISE image the variations are insignificant. The problem only really arises for images or mosaics covering a larger region.

Phil

Fully agree with you that with most projections the variations are very very small, but you never know with these rovers, by the time we start mapping the 100 km drive after Endeavour we have to take care

Haven't done much more but I decided to pack up what I have done and let people play with it.

At the following URL you can download a simple app that I have done to plot routes using red/blue/green terrain maps. I have included Mikes terrain map for convenience and to save using Doug's bandwidth

http://tinyurl.com/5k3tc8

It is an executable java JAR file so people should have no trouble hopefully. You will need java version 1.5 or greater though (safest would be 1.6).

When you run the JAR file just go File->Open Image.
Chose a terrain image like the one I provide for download. I designed it using RGB values (red = bad, green = good, blue = medium)

The constants used to indicate the cost of crossing a particular colour can be changed by entering the values on top
As discussed above:

The cost function is

If you change a value, be sure to hit the 'enter' key before moving on.
To indicate a start and end point just click on the map and hit 'Start' You can indicate as many points as you like. It just traverses from the first to the last. The size of the grid used to break up the terrain map can be adjusted from the 'Grid Location Size' menu. The estimator menu changes the method used for estimating the distance between points for the A* algorithm.

To clear all your points hit the 'Clear' button.

Finally you can save your maps using the 'Save Image' button. Just make sure you put in the full file name as a PNG image when saving, e.g myMap.png

That's it I think. If there are problems I'll do my best to change them at some point.

-Ed

Edit: Note there may be a bug in what I have distributed. The behaviour when all weights are 0 seems off.

Thanks jaredGalen, it runs perfectly under Linux... once I'd worked out that I had to click on the image a couple of times to define start and end points before hitting 'Start', if I wanted anything to happen, though. D'oh!

The temptation to try it out on absurd images is great, though. Hmmm, how DO you get to the bottom of Valles Marineris, anyway? ... *cough* ... also, I note that it throws "java.lang.OutOfMemoryError: Java heap space" on STDERR when given a ridiculously hi-res image, though that doesn't show up in the GUI.

Yeah, I guess good instructions were scarce in my post You need at least 2 points for it to work, start and end point.


And yep, java.lang.OutOfMemoryError can happen, I handle it wonderfully by ignoring it.
If the terminal is your favourite place then 'java -jar -Xmx1024m RoutePlanner.jar' is your friend, give it as much as you can and see how big an image it can take

All in all I didn't try and account for much user jiggery pokkery, so it will run into trouble. And also, I hate doing GUI stuff
Still haven't checked out my potential bug too

Thanks for sharing!

-Ed

Computer-generated route using directional STDEV, from position at sol 1687 to the point where Mike's SW route and W-spur routes join up with the others. This would require a turn to the southeast which as yet doesn't seem likely.

Click to view attachment

Total route with STDEV values for southerly course:

Click to view attachment

latest version (2) of my toolkit can be installed from here

Based on Eduardo Tesheiner's Opportunity Route Map of Sol 1687 (now solidly on all five UMSF Terrain models):

HiRise image with W Spur route (on left) and SW Passage route (on right) indicated:
Click to view attachment

Shift Differential Terrain Model with W Spur route (on left) and SW Passage route (indicated on right):
Click to view attachment

Looks like Oppy is headed toward the W Spur route entrance.

-Mike

All of the UMSF Terrain Models with Oppy's Sol 1687 position indicated:
Click to view attachment

(BTW, let me know what y'all think of this format - it's kind of a pain to put together.)
-Mike

The comparision is very informative Mike.

> Looks like Oppy is headed toward the W Spur route entrance.

Agree, although I have some discrepancies about the exact entrance point because there are some big ripples right there.

Full Hires STDEV analyzes centered on sol 1687 position with past track in yellow and West Spur route marked in orange.

Click to view attachment

I agree with Tesheiner that the entrance point of the W-Spur route is a bit uncertain, also from the point of moving towards Endeavour the W-Spur still doesn't make sense to me mathematically, if I try to compile an optimal route, based on Mike's driving-assesments, the software always seems to favour the SW route as the fastest route from the 1687 position (see also my earlier post).

I still expect they intend to turn south or SSE later and head for the SW route, although that one also has some nasty spots it gets you quicker across the debris zone and enroute to Endeavour. The W-Spur route would only make sense if you lose a wheel, or there might be scientific reasons to head for it.

Next driving days will be interesting..

FYI - Here is Tesheiner's route map with my colour coding. Dangerous seas ahead...

This is my guess for the route to come.

Phil

Click to view attachment

There a couple things ahead that look like they might be mini-Erebus-style dust traps. These seem to have a circular shape with a smoother interior (more green color), a rock pavement rim, and in the HiRise images they seem to be just slightly darker than the surrounding sands.

They show up nicely in my terrain model probably due to the color shading levels. (They also show up in Fran Ontanaya's terrain model).

I've indicated a few of the more obvious ones with dark arrows (there are more in there...) in the area Oppy is about to enter in this blink animation:.
Click to view attachment
(click to animate)

Digging back through Oppy's historical route, I didn't see any similar type terrain patterns that Oppy passed through. (There were a few close by to Oppy's route but I couldn't find any corresponding Navcam images). So I have no idea if this is a real concern or not...

-Mike

STDEV analyses for SW driving direction with actual traject (yellow) till sol 1691 and route proposed by Phil Stooke (orange).

I agree that this indeed looks like the most likely route Oppy will take and it looks like the smoothest route possible (although not necessarily the quickest), but Mike is correct that there are some features there which require attention, maybe old craters filled with dust or something similar. I wish Paolo and his colleagues good luck on this route, and keep your eyes open ;-)

Click to view attachment

After the final turn to SE on the route there might be some nasty parts, little bit too much red in the picture although there doesn't seem to be much choice... only possible option might be to continue SW or WSW, keeping to the west of the big red area's and then turning a bit more south to get to the green, this is a slightly longer route but avoids the nasty red patches after the turn to SE.

Not a new analysis here, but a new dimension to support (I hope) other analyses.

A red-blue anaglyph showing our probable future route, the so called West Spur, and bedrock to the east. I have put it into 25-% perspective (not geometrically correct) for the effect and to save space. The original images are half resolution of the HiRise images (the scale 0.5 m/pix is retained in the bottom in E-W direction). Oppy's current location is at the upper right corner.

In the southwest there seems to be an alarmingly deep eroded crater, the big smooth area in the southern part of the West Spur route. Might be wiser to skirt around it?

Very cool! (nice pun, too!) How did you create the anaglyph?

Seems that I'm after all going to analyse... To put things still more into perspective (maybe a pun too), here's comparison of Erebus (top) and the above mentioned southern ghost crater presented in a similar fashion, in the same scale (here 75 % of full-res in E-W dir). The latter crater being very eroded, it's a bit hard to judge its dimensions, but I'm getting the idea that it's slightly less deep of the two though.

I have used the HiRISE images PSP_001414_1780 and PSP_009141_1780 (using OpenEV to export samples, StereoPhotoMaker to make anaglyphs). They were acquired at one Martian year interval and have a very similar lighting and shading. And they make a spectacular stereo pair, actually I think the effect is rather exaggerated, but very suitable for this topography.

My using of the (pseudo) perspective as shown is just an idea meant to more easily handle and visualize these images while at the same time to keep the resolution in the E-W direction as good as possible, obviously of central importance in this case.

I'll perhaps post some more (full-res or larger) anaglyphs after Oppy has chosen her way between the ripples.

That is a very interesting effect. Smoother terrain seems to be at a lower perspective depth.
How does it treat the flat terrain just north of Erebus crater? (Seen in Sol 624, here)

Do you think it would be possible to use the perspective view of the two acquired images to create a "flat" 2D terrain model (with coloring based on perspective depth)?
(IIRC, this type of analysis is actually closer to what the rover drivers are using.)

-Mike

Essentially you would be creating a DEM from the data, something that isn't exactly very easy to do. While our eyes can easily determine the "distance" between a point in two stereo images, a computer can't very easily do the same. However, various approximations can be used that would be perfectly acceptable here. With any of these (most have to do with STDEV or variance analysis between the images) one quickly runs into an issue in that the overall elevation of the land can vary significantly across the image, causing both incorrect colors and incorrect matching between images. Applying a high-pass filter should do the trick, though. This is what I experimented with way back here. However, then I did not have the capability to easily extract large areas of the image and work with them in PS/GIMP or PIL/ImageMagick.

One last image at JuraMike's request. (It's pretty daunting work to make custom anaglyphs in this way.)

The stretch from Viking & Voyager to Erebus north. Purgatory is visible in between. This time full res scale E-W (in the image bottom) and about 40 % scale lengthwise, again with an arbitrary perspective to reduce size.

A problem rendering the images less usable for DEMs is apparent: a slanting 'slash' through the image center that appears to shift the left side 'higher', or nearer to viewer, than the right side. In the otherwise pretty flat portion of the landscape this is quite striking feature. I think it's an artefact related to the map projection.

I'm wondering if it could be done semi-manually:

1)The red-blue patterns get separated into two greyscale images.
2) The modeler determines which is going to be the "back" part of the image and lines up the images so there is minimal coordinate change for the back part. (This is where the modeller also picks a "typical" terrain in the image.)
3) An image combination (multiply, darken, lighten, invert one then do stuff??) is made that highlights the differences.
4) The whole thing gets Gaussian blurred (and rescaled, contrast enhanced, colorized, etc.)
5) Overlay onto HiRise hi-res image.

This would be a real similar process to the shift-differential recipe BUT it uses two separate HiRise images from different angles to get topographic variation rather than just a pattern shift variation.

Might give new information that could be used alone or combined with one or all the other terrain models.

-Mike

This is getting exciting. One thing that is NOT apparent is the smooth flat area that Oppy crossed on Sol 624. It doesn't look like a dust trap in this image, nor did it look like a dust trap in Oppy's Navcam image. (I'm thinking this technique might be helpful for spotting slightly lower depressions that could hold poofy dust.)

The next few Sol's images from Oppy can be compared to your earlier images of the W Spur route region.

Was this area also modeled by jekbradbury?

-Mike

Sol 1691 Predicted vs. Actual from the Shift-Differential Terrain model (pretty good!) using a crop of Ant103's image:
Click to view attachment

I'll try to do a better comparison when the Navcam and Pancam images (for soil type comparisons) get loaded onto the MER Opportunity homepages.

-Mike

Mike, sol 412 pancams could match http://marsrovers.jpl.nasa.gov/gallery/all...unity_p412.html

Good call, Tman!

Here is the view looking forward and backward from Sol 412 (which was also a long drive day) compared with the Shift Differential Terrain Model:
Click to view attachment

(Just like back then, it looks like we are about to enter increasingly wider dunes)

-Mike

Sol 1691 image vs. Sol 412 image. (this is not supposed to be a stereo image)
Click to view attachment

Spooky!

-Mike

Crudely colorized rear hazcams from last sol.

Click to view attachment

The sand holds well under those wheels.

I need some advice from the UMSF image-mage brain trust.

I'm looking for a way to normalize and then compare the various UMSF-created Terrain Models.
They are all currently in a green-blue-red color scale and I'd like to convert them all into a grayscale.
So what I need is a reverse color table.

Any advice on how to do this?
(I use Photoshop CS3)

-Mike

Thanks for the suggestions!

I used a Black and White adjustment layer for the color images and fiddled with the various slider bars until it matched up with a pre-colorized version of the Diff-Shift grayscale image.

It seems that each color image will have to be manually fiddled with to match up to a standard. (I was kinda hoping for a more systematic approach.)
The process ain't pretty, but it works!

Here is a blink comparison with the Shift-Differential Terrain Model and the FT Terrain Model for the area that Oppy is about to traverse.
Click to view attachment
(click to animate)

I'd suspected that these were going to be similar, and it seems they can be matched up pretty well. I don't see any obvious features that are indicated in one model and not the other.

(Due to the 30 pixel Gaussian blur applied to the Shift-Differential model, the crater floors of smaller craters come out darker and clearer in the FT model. But I'll wager a prudent driver would want to avoid these or at least approach with caution!)

-Mike

5-Color STDEV analyzes (multidirectional) of terrain immediately ahead

Click to view attachment

and same image and same analyze using grayscale instead of RGB colors:

Click to view attachment

Looking at IASViewer at highest resolution it looks like, in a southerly direction, the dunes only get wider for a short while, afterwards they get smaller and more nasty looking. Personally I don't see a clear 'highway' on bedrock dead ahead, there might be a few lanes but all of them get closed pretty quickly by cross dunes. There is good terrain to the southeast, but then I wonder why they didn't turn towards that entrance earlier. To me Phil Stookes route still looks the most smooth route, with your SW-route the quickest (even though it has to cross some nasty terrain later on). But I'm not a driver and they've probably got better data then we have...

Click to view attachment

This is an STDEV analyzes at 50 cm/pixel for a southerly driving direction with rover track showing in yellow. As can be seen the few open lanes dead ahead all of them close quickly with red, just does not look so good, however we have to take in mind that all of these analyze methods don't work so good on bedrock. It gets really interesting where our real drivers are setting course to...

Great image Fran Ontanaya ... the sky looks pretty pinkish in color due to the dust in the atmosphere I guess ?

@PhilCo126: I think it's early in the morning in the 10.26 and 10.29 hazcams; anyway, that image is colorized from B&W.

Actual [Sol 1693] vs. Predicted [like Sol 582] (Blue terrain looking towards Magenta):
Click to view attachment

Hard to get a sense of scale in the Sol 1693 image without Oppy in the foreground, but I think the dune wavelength is a match!

-Mike

Predicted computer-route from sol 1695 onwards (orange line), the computer avoids the biggest dunes and then steers more to the southeast towards clear ground.

Click to view attachment

OK, time to lay the marker down...

Predicted view of what the Sol 1699 Navcam View forward will look like:
Click to view attachment

I'm projecting the heading will track ca. 50 m drives following the 1693-1695 heading (SSW). This will put Oppy on Magenta Terrain on Pavement in the half-pipe dune direction. So it should look like similar terrain to Sol 823.

-Mike

Taking the Shift-Differential model as having the standard color range, I coordinated and normalized all the UMSF Terrain models as grayscale.

Here is a rough draft blink animation that cycles through all five grayscale UMSF terrain models:
Click to view attachment
(click to animate)

-Mike

Sol 1697 Oppy Position Coordinated, Normalized, AND Re-Colorized Terrain Models:
Click to view attachment

-Mike

... and a nice green coloured patch to the right; the SW corridor. Sigh.

Given that it's obvious there are bigger dunes in this area even without the fantastic dune-height maps, I'm thinking they're just testing some autonav capabilities. It will be interesting to see if they shift to better terrain after trying things out a bit.

Here are TIF files of all the coordinated, normalized, and grayscaled UMSF Terrain Models. They are at 12.5% of the full resolution, (so they are at 2 m/pixel). Each file is about 10 Mb.

They should line up automatically if you stack them all together:

S Victoria HiRise image grayscale (coordination basemap): http://www.speedyshare.com/350935464.html
S Victoria Malaska 20081002 Terrain Model coordinated and normalized grayscale: http://www.speedyshare.com/542835747.html
S Victoria Canvin 20081001 Terrain Model coordinated and normalized grayscale: http://www.speedyshare.com/930855894.html
S Victoria Ontanaya 20081001 Terrain Model coordinated and normalized grayscale: http://www.speedyshare.com/901693914.html
S Victoria Butler 20081006 Terrain model coordinated and normalized grayscale: http://www.speedyshare.com/569264944.html
S Victoria Sassen 20081031 Terrain model coordinated and normalized grayscale: http://www.speedyshare.com/477729197.html

Enjoy!

-Mike

Side-by-side comparison of the normalized grayscaled UMSF Terrain Models of the S Victoria Region:
Click to view attachment

-Mike

Side-by-side comparison of the the normalized GBR colorized UMSF Terrain models of the S Victoria Region:
Click to view attachment

-Mike

Here are TIF files of all the coordinated, normalized, and GRB colorized UMSF Terrain Models. They are at 12.5% of the full resolution, (= 2 m/pixel). Each file is about 30 Mb.

They should line up automatically if you stack them all together:


S Victoria Malaska 20081002 Terrain Model coordinated and normalized GBR colorized: http://www.speedyshare.com/991410266.html
S Victoria Canvin 20081001 Terrain Model coordinated and normalized GBR colorized: http://www.speedyshare.com/822224262.html
S Victoria Ontanaya 20081001 Terrain Model coordinated and normalized GBR colorized: http://www.speedyshare.com/664314485.html
S Victoria Butler 20081006 Terrain model coordinated and normalized GBR colorized: http://www.speedyshare.com/813536658.html
S Victoria Sassen 20081031 Terrain model coordinated and normalized GBR colorized: http://www.speedyshare.com/620227899.html
(basemap): S Victoria HiRise image RED-->Grayscale (coordination basemap): http://www.speedyshare.com/641932676.html

Enjoy!

-Mike

They drove a little farther than predicted. This puts Oppy on Magenta terrain in dunes. Probably more like Sol 853 than Sol 823

Original prediction compared to actual (original position would have Oppy in Magenta terrain on pavement):
Click to view attachment

Modified prediction (Oppy on more duney magenta terrain):
Click to view attachment

Taking into account the view angle, the dune heights and half-pipe widths seem to match!
[cue triumphant air guitar solo]

-Mike

Double post deleted. Mod.

Are these UMSF terrain models actually doing anything, or are they just highlighting subtle lighting differences?

I took the original HiRise image, and normalized it (stretched it) to fit the Shift Differential 20081003 Terrain Model grayscale range.
Then GBR colorized it.

Here is a comparison of the three versions of the HiRise image [normal, grayscale stretched, and GBR colorized] along with the Shift Differential terrain model [grayscale and GBR colorized].
Click to view attachment


There are some regions that appear different in the two images. In particular, the area just to the W of center of the image. (At the intersection of a line WSW of "the Porcupine" and a line S of the center of Victoria Crater apron).

Another area is the region directly S of Victoria at the edge of the dark apron. This area was lit up as "nasty bad red" by the Terrain models, but seems more reasonable in the HiRise images.

Here are links to the 12.5% resolution "normalized" (stretched) HiRise images:
S Victoria HiRise image normalized (stretched) grayscale (10 Mb TIF file): http://www.speedyshare.com/151690732.html
S Victoria HiRise image normalized (stretched) GRB colorized (30 Mb TIF file): http://www.speedyshare.com/991145238.html

-Mike

I think what you're highlighting here is that smooth terrain tends to be dark, and ripples increase the average brightness. We discussed this earlier in the thread - my concern was that if you used this simple "lightness" method you would be confused by smooth areas that were brighter than others, eg.

What's the current thinking on ripples on bedrock? I was concerned that, at least with variance or FT methods, the brighter background of bedrock makes the lightness fluctuations much larger on bedrock than on sand, for ripples of the same size. Does this confuse the current methods, and give drivability ratings that are worse than they should be over bedrock?

I think all the methods are looking at the wavelength of the alternating bright-dark patterns (I can't speak for Fran's method).

For the shift-differential method, I manually selected the bedrock areas and brought it down to a "neutral" background value (160/256).
I hoped that for the differential part of it, the subtraction of like neighboring values should make regional brightness variations cancel out.

For the FT methods, in theory, I think the bright-dark pattern of pavement alternating with dune sands at a wide wavelength could fool the model into thinking the area was more hazardous than it really is. In practice, the reason that the pavement is broken up into bands is because they are indeed obscured by wide wavelength dunes. And these big wavelength dunes on rock are still scary, if you try to cross them or hit a pinch-out.

If however, there was an edge-on rock layering visible at the surface and it was orthogonal to the FT or shift-differential directions, the FT methods (or the differential shift method) might not work as the brightness variations might fool the modeling methods into thinking there are big dunes down there.

(Kinda trying to think about this, I'm postulating that the slicker smoother surfaces of the flat pavement help the intervening dunes pile up a little bit more than if they were on sandy terrain. So the area S of Victoria should be a regionally uplifted zone of rock, now covered by alternating dune bands. If the rock should ever get covered with a layer of sand that doesn't get removed, the dunes might flatten out and you'd get a smoother area. *I think*.)

What I don't understand the smooth region of dark sands immediately around Victoria Crater: It's topographically higher and more exposed to the wind, why isn't it all bare pavement?

-Mike

Good question. One guess is eons of deposition of dust/sand ground from the cliffs and blown out on the winds. We know this is happening with the dark streaks to the north of Victoria. With varying wind directions over the millenia I could imagine the entire annulus coated this way.

I raised a similar question earlier about why are the ripple fields apparently associated with Victoria. Are the ripples derived from stuff that blew out of Victoria? Why are there no ripples very close to Victoria?

If I read Parker's comments correctly, the new hirise image shows clear sailing once we cross the current ripple field. It would be interesting to apply the FT/variance/shift/etc techniques (or just the good 'ol fashioned eyeball!) to large regions of Meridiani covered by hirise, to see if the ripple fields really are associated with Victoria-type craters.