wow



wow

Long time lurker, first time poster.

I played around with a texture classifier I have access to at work. I picked a few image attributes I thought might be relevant, and nominated a few regions of example textures into which I wanted the image classified. The two attached images are from a place near the destination, with the classification in color. I thought it nicely segmented out the difficult ripple terrain from the rest. The polygons you see in the images are some of the nominated regions. Lots of things can be tweaked, like the choice of attributes, the example polygons, and the colors. I think I picked too many example areas, since the good areas are a mixture of colors, but the rippled areas area clearly segmented out.
The area is near the destination of Edurance, and the image has been rotated - just an artifact of my importation to the processing system that I forgot to undo.

I have also tried doing 2d FFTs to extract ripple spacing, orientation, and strength, from the maxima of the spectra. This seems successful, and I'll post some of those images shortly.
With these FFT attributes available to the texture classifier, it might be possible to nominate even the different ripple styles and have them individually classified. I have not tried this yet.

PS Since this is my first post, I'm not sure if the images will come through OK. If it doesn't, I'll try again!

Bill Butler

Followup to last post. These images are the raw image I started with, and the standard deviation attribute, similar to what some others were playing with. I think this attribute is a good indication of ripple height/width, which seem to me pretty well correlated with each other.

Bill

And my last post for the moment - The three 2DFFT attributes. I look for the maximal fft value, other than at the origin, and convert to radial coordinates to compute a distance and angle in frequency space. I take the reciprocal of the frequency distance to get a real distance, and I wasn't sure what to do with the angle - I think I can pretty much leave it alone (is there a rotation or something?). I also plot the absolute value of that maximal fft value. This should be related to the strength of the ripples, but will also be related to their regularity. Strong ripples with varying periodicity will spread out the energy over several frequencies, so the maximum will be lower. I think the standard deviation attribute does a better job of this. On the other hand, if both attributes are given to a classifier, maybe it could distinguish strong regular ripples from more chaotic ones.

The titles of the images are in the upper left hand legend in the picture. Their meanings are:

rippledist - reciprocal of the frequency distance. Should be correlated with ripple spacing regardless of orientation. Not calibrated to any units.
ripplearg - angle of ripple (in frequency space, but should be something very close to physical space)
rippleamp - the absolute value of the maximal FFT sample

BTW for these images I did compte every pixel, as the subimage I was using was 2000x1100. WIth a 64x64 fft, this took about 5-10 minutes. Doing every third pixel goes 9x faster and I think give virtually as good information. For the full image, it may even require decimating further, but I could also imagine focusing on the likely rover paths and doing more detail.

I see I've run out of space for images for this post. I'll put the last one in the next post. If there is a better way to post these images, let me know. Thanks!

Bill

And the last image:

Welcome, Bill!

This is fantastic stuff! Would it be possible to apply this to the terrain area near Erebus and the etched terrain? That area seems some of the gnarliest and it'd be great to do another side-by-side comparisons of the recent generation of methods.

I'd really like to see if your technique lights up the bad/scary/curvy dunes. (I think/hope it might)

-Mike

EXCESS QUOTING REMOVED - ADMIN

Great stuff Bill!

My tool works as yet only with the standard deviation, calculating the standard deviation in a 360 degree circle around each pixel, using a range and sample rate which can be configured. I improved a bit on the tool to make it easier to work with and will add some more improvements as I go along, but this standard deviation trick seems to pick out the 'bad' terrain reasonably well. On the other hand, Fourier will get a better indication of the actual 'ripple wavelength', however this only works accurately along the direction of the ripples. I have the feeling that running a FFT along 360 degrees will give you a meaningless answer or at the most something identical to the standard deviation trick.


Click to view attachment

Terrain roughness as indicated by standard deviation goes from green to yellow to orange to red to purple with increasing value (purple is worst). In above example the open bedrock areas are still shown purple, however a lot depends on how you set the resolution, a higher resolution results in a lot more time for calculation, but shows finer details, if I run part of the picture on high resolution the bedrock shows orange or even yellow:

Click to view attachment

What I am working on now at the moment is to see if I can make the thing show the general direction of the ripples across the image, and I am trying if I somehow can make a distinction in 'driving direction', in other words big long ripples might be okay to drive as long as you remain in the troughs between them (and as long as these troughs remain open for long distances, no cross-ripples), while they are bad if you need to drive across them. So, as we know the 'required direction' to the destination from every position in the picture, it is possible to see how this translates in terrain classification. Once I get more results of this, I'll show them.

Just to get an impression where we are at the moment, maybe it would be worthwhile if somebody posted one image which we can then all analyze with the various tools, each tool will probably tell a different story but that only makes it more interesting as every trick seems to indicate one specific part and if we put them all together (preferably on a picture showing route already traveled) it is easier to see which tool tells us what. Note it is definitely not a competition to see which trick is 'best' as each has its own specialization, but if we for instance find a route which is voted 'green' in all analyzing methods, that's perfect, the less votes a route gets, the more the 'real' drivers have to take care. Finally, in the end, it all comes down to human eyes, but the tools will help for a first classification of the terrain I guess.

Regards,

Geert.

I've been using the Erebus to Etched terrain image.

Available here (November 29 MRO Image Release thread, Post 33)

-Mike

I wonder if this enthralling study of MER route prognostication should be ensconced in its own thread, rather than interspersed with routine progress reports for Oppy? Los Moderates?

Whoo-hoo! The differential-shift method is working!

Following Circum's suggestion, the key is to remove the brighter-toned bedrock and replace with a slightly darker gray (color tone of sl. shaded dunes) that average out with remnant bright edges of bedrock. This averages out after Gaussian to make the bedrock cancel out during the differential shift.

(I played with shifting values and Gaussian blur to see which shift gave the brightest response for big dunes with a 22 pixel wavelength. Before blurring the brightest post-differential response was a 15 pixel shift to the E. After a 4 pixel Gaussian blur, the brightest response was 10 pixel. The reason is that the dunes are not symmetrical, but the Gaussian blur smooths them out enough.)

Bedrock areas with big dunes are also getting indicated as "bad". But bedrock areas with small dunes are getting flagged as "good".

Here is the result:
Click to view attachment

And here is the recipe (big dune wavelength about 22 pixels):
1) Select the brightest bedrock areas (I cheated and used the magic wand function with a tolerance of "10", it gets a few dune tops but it's OK)
2) Recolorize to a darker dune-shade gray
3) Copy layer
4) Gaussian layer each layer (4 pixels, about 0.2 dune wavelengths)
5) Shift one layer to E (orthogonal to parallel dunes) by 0.5 dune wavelengths (10 pixels)
6) Take differential
7) Merge layers
8) Gaussian blur to 30 pixels (1.5 dune wavelengths)
9) Contrast levels. Darkest = black, brightest = white
10) Colorize green=black, red = white
11) Overlay (multiply) with original image.

-Mike

Side-by-side comparison of the Erebus Crater to Etched Terrain images done today compared with Oppy's historic route:
Click to view attachment

The models are converging!

-Mike

[EDIT 20080927, 01:42 EDT]: Caption for Fran's image corrected]

Uh, I'm not using brightness anymore. It's a home cooked sort of edge detection (yeah, I realized it after doing it).

The previous map without the dune throughs was actually showing "edge density" of areas (by blurring the detected edges), so it was a poor man's multidirectional 'frequency analysis' method. It works for the bedrock because here it is cracked.

Thanks for the side-by-sides, Mike. I find them very helpful. The Erebus south segment is an excellent test area.
That is really looking very nice, James. I can't wait to see it in full resolution.

Whoa! That is very interesting, and it confirms my earlier comment that useful terrain classification software already exists. It has been my experience that industry has expertise and technology in some areas that is every bit as good as as that used by government agencies and academia, and often, even better. I'd really like to see what you can do with the drive segment Juramike has suggested for comparisons...from Erebus to the etched terrain. Is it possible that your company would be willing to make their proprietary software available to the mission?

My version of the same picture, using standard deviation you can find here:

http://www.navtools.nl/roverrouter/erebustoedgedGeert.jpg

Purple is worst, orange best. I'll have to fiddle a bit more with the color settings as red is a bit overdone maybe but it looks like it agrees with the others.

As mentioned before, I'm trying to sharpen down the various definitions, some place can be 'bad' if you proceed in one direction, but 'reasonable' if you proceed in an other direction, I think I can make this visible (if I can find the time ;-) ).

Regards,

Geert.

Yikes, that is a large picture to post in the thread. I think it would be better to make it an attachment.

We should keep this thread for the terrain analysis.
Moved a few post here: Leaving Victoria crater

Well I wasn't planning on making maps at the full HiRISE resolution, it is just too big. The 1/4 version is enough to see the ripples. I may do a few sections at higher resolution for the area just south of Victoria that we are likely to be crossing soon, but for general route planning I think this scale (1m/pixel) is better.

Of course I've posted my raw FT map at it's full resolution (8m/pixel) and you have the full resolution HiRISE, so if anyone wants to play with huge images then go ahead.



It's a matter of how you interpret the map. It's not necessarily red=bad, green=good. Red should be fine if you stick to the troughs, Oppy got through some of the worst areas already! It's more "you'll be limited in the direction of travel and maneuverability" rather than "bad" So 1) in red/orange areas Oppy will probably be limited to heading south 2) If she encounters any obstacles (large ripples, dead end trough, etc.) it will be harder to go around them.

I don't see the point of mapping every ripple - they are obvious in the raw HiRISE alone.

James

I see this as: red - you must have a look to the full res HiRISE image to plan a drive, and green - you may "blind drive".

Yup. I'm going to try to incorporate this color scheme:

Green - Parking lot. Blind drive any direction OK (like terrain just outside Eagle Crater)

Yellow - Some small dune crossings - Cross-dune blind driving probably OK

Orange - Larger dunes - Dune crossings or cross-dune could be problematic. Route adjustment during dune crossings necessary.

Red - Large dunes - driving in troughs may be OK, but dune crossings or half-pipe pinch-outs could be serious trouble. Careful drive planning will be necessary.


Rating a few key sections Oppy's route from Erebus to Etched Terrain based on Oppy's images would be a good thing that would help coordinate the orbital image processing with the ground truth - but I doubt I'll have time to do it anytime soon....

-Mike

Fran, your map is exactly what I was thinking could be done. There is the need to classify on a large scale (parking lot vs difficult), but we will have to cross difficult areas, and dune level fine scale classification can be helpful in path finding through these areas. The HiRISE images are excellent, but processing them to show slope brings out really critical information.

One attempt to make the calculation 'direction-aware', instead of a 360-degree check on standard deviation it now shows only the standard deviation in the direction of a set 'target' (shown as a cross in the image). The picture will change significantly if the target is moved across the screen. Colors once again state degree of 'roughness' as indicated by standard deviation. Given the direction of the ripples, usually positions east and west of the target are showing 'bad' and positions north and south of the target are showing far more favorable, however the interesting part is that there do remain some 'roads' to approach the target even from east or west.
Click to view attachment

Second picture is a first attempt to calculate ripple directions and wavelengths, the arrows in the picture show calculated directions, and the colors revert to the wavelength (purple is the longest, then down to red and orange). Generally it seems to work, however the formula I used is still prone to confusion with shadows from other ripples, cross-ripples, etc, this will need some fine-tuning but eventually it might be worthwhile to do.
Click to view attachment

The tool I use for this can be downloaded for free from http://www.navtools.nl/roverrouter.htm for those of you who use MS Windows XP/Vista.

Now I've convinced myself that the Fourier method works it's time to look ahead.

Here is the 'difficult bit' Between Victoria and the green pastures beyond. Including a stupidly large full resolution version.



Looks like there are a few paths through.

James

I have applied my tools to the erebus image to make comparisons easier. I have five new images.
Three of them (amp, dist, ang) are based on a 64x64 fft. I find the sample in the 2d fft plane with the maximal absolute value, then convert to radial coordinates and invert the distance to get a real distance. I do the calculation on every third sample in both directions, and put the result in all 9 pixels around the target sample.

They are:

amp - the absolute value of the largest fft sample.
ang - the angular component in degrees of the largest fft sample. Correlates well with ripple direction
dist - ripple spacing, in arbitrary units, from the radial component fft space
mean, stddev - computed on the same 64x64 moving panel as the ffts.

In my opinion, the dist attribute is very useful, as is the mean and stddev (although these appear well correlated with each other). The ang attribute should not be useful for traversability, except when you only want to consider traverses in a particular direction. Even so, the ripples are so uniformly oriented I don't think we need this one.

Ultimately, I will try the texture classifier with these attributes and possibly some others, attempting something along the lines of Juramike's suggestion. You can see my nominated regions, which I call bedrock, small, large, huge ripples. There is no parking lot texture in this image. Any suggestions about the location of these nominations are welcome.

Bill

I tried to classify Oppy's path around Erebus Crater based on drive tracks across dunes. Most of the route was south along the troughs, so I also tried to use tracks across the occasional dune crest as well as views towards dunes to estimate difficulty.

Navcam images included:

Click to view attachmentClick to view attachmentClick to view attachment

More images (the smooth terrain NW or Erebus, and working around the W side of Erebus Crater);

Click to view attachmentClick to view attachmentClick to view attachment

Now working towards the Etched Terrain.
In some cases I added the Navcam view looking directly towards the dunes to try to estimate their waviness.
Click to view attachmentClick to view attachmentClick to view attachment

Sol 778 and 780 classifications and Navcam images:
Click to view attachmentClick to view attachment

And putting it all together and overlaying on the Erebus Crater to Etched Terrain test HiRise image:
Click to view attachment

I'll use this for the "Ground Truth" to compare all the next generation of processing techniques.

-Mike

Anybody know if there's are links available to hi-res HiRise jpegs of the areas between Victoria and Endeavour?

A little more tweaking and I'll be ready to play....

-Mike

Here's a crazy, far-fetched idea that just might work: Most of the area in question (Victoria to Mini Endurance) is covered by not one, but two HiRISE images. Chances are, they were taken from slightly different orbital positions, so overlaying them, taking the difference, and applying a Fourier analysis to what is essentially a high-resolution map of small-scale elevation changes should give a result that actually reflects what is the toughest part of these dunes: their height, not their length or width. All this is just a guess, though. First, everyone could try remapping their color gradients to Juramike's ground truth color values.

On hires JPEGs, that's exactly what I'm looking for, too. Maybe Emily will be nice enough to provide them again?

Not a crazy idea at all. DEMs Digital elevation models) from those multiple passes have been computed, and it is not at all a trivial matter. Unfortunately the best that can be done in elevation is an accuracy of about one foot. That is unfortunately insufficient for reliably finding a good spot to cross ripples or finding their size. If they are 1 meter tall you can immediately tell from visual (or automated) inspection of the HiRISE image.

Paolo

Wow, I can't believe you actually went through the trouble of doing this. I DO NOT want to discourage you (or anybody else for that matter) from doing this, but I will have someone at JPL do this analysis because they will be able to also look at the telemetry (I really wish I could involve you in this!). So many sols have gone by and we need to refresh our memory.

You guys never cease to amaze me.

Paolo

The "DEM" method has some promise:

Click to view attachment

Instead of trying to get a six-inch accuracy map of each dune, it simply looks at the average "height" within a 25-pixel radius and compares it to the average "height" within a 100-pixel radius. This is because the elevation actually changes significantly across the image (it's the edge of the ejecta blanket) and would mess up the colors.

Are the DEM's accurate enough to detect the change in elevation mentioned here
http://marsrovers.jpl.nasa.gov/gallery/pre.../20050506b.html

I'm still concerned about the fish scale pattern beyond the etched terrain south of Victoria representing larger scale drifts underlying the more obvious ripples. The drift Oppy got stuck in at Purgatory was on top of one of these, see the attachment to Pando's post made at the time.
http://www.unmannedspaceflight.com/index.p...ost&p=11840

Not really. But now, after purgatory we have a new FSW (flight software) that can alert the rover it is potentially stuck in sand. When Opportunity was in what you guys have called the "Quackmire" the rover successfully detected it was stuck. That was sort of unfortunate since we were trying to get unstuck . Therefore, even in the case the rover slams in one of these purgatory stye ripples the rover should stop before it gets too deep into trouble.


You are correct in identifying the "fish scale pattern" with purgatory-style ripples. Fortunately they are clearly visible in HiRISE. If we are able to precisely localize the rover in that area, it might slow down the per-sol progress, but it should not be considered a threat to the rover. I'm moderately optimistic.

Paolo

I have been watching this thread with great interest and little time.

But then it happened. My 'honey do list' were canceled due to a hurricane warning (Cat 1 Kyle) approaching the area. We are all battened down (outside Boston MA) and I had a few hours to devote to this task.

I am using a quick and dirty DEM approach. Each one-dimensional strip crosses the 'mostly' north-south oriented dunes. Using the brightness and knowledge of the Sun angle I can keep a running pixel-by-pixel tab (relative) of the elevation. The bedrock is the flat reference and they tend to be flat for many pixels (compared to dune peaks). Long brightening dune faces give me 'height' and the darkened sides(shadow) give me the height down. I then plot a running average of height and find the 'driveable spaces' between dunes.

I am aware that there are field of landmines with the technique. and I am only on Version 0.001

here is a chunk of the full res Erebus chunk (thanks Emily) and the link below is to flikr (but since I dont have a PRO account it is severely shrunk...sorry)

In the full rez chunk, you can see Oppy's tracks... (this chunk) is just north and heading into Erebus.
There is also an annoying diagonal line artifact (I have a one off C pointer somewhere).

Green is drivable
Dark and red is not
White is bedrock (and very bright dunes....Version 0.002 is comin')

(comment: my map indicates there is a always a drivable 'dune gully' somewhere close. Its just a matter of planning the way forward...like a maze...)

Click to view attachment


here is flickr link:

http://www.flickr.com/photos/30904379@N07

Cheers! (wow...the rain is coming down sideways!)

I see most of the algorithms analyzing the terrain aren't flagging them as much as the terrain between Erebus and Victoria, not sure I should see that as good or bad. Hopefully I'm just being paranoid about them, looking back Oppy did get through a significant stretch of the before getting stuck in the last one in her path

Nice work you all but, you'd better hurry :
http://www.unmannedspaceflight.com/index.p...mp;#entry127004
see post #14

Changed the name of this thread to reflect what it's really about

And the next thread we can call "On the Road Again"

On the road again
Just can't wait to get on the road again
The life I love is makin' music with my friends
And I can't wait to get on the road again
On the road again
Goin' places that I've never been
Seein' things that I may never see again,
And I can't wait to get on the road again.

On the road again
Like a band of gypsies we go down the highway
We're the best of friends
Insisting that the world be turnin' our way
And our way
Is on the road again
Just can't wait to get on the road again
The life I love is makin' music with my friends
And I can't wait to get on the road again

On the road again
Like a band of gypsies we go down the highway
We're the best of friends
Insisting that the world be turnin' our way
And our way
Is on the road again

Just can't wait to get on the road again
The life I love is makin' music with my friends
And I can't wait to get on the road again
And I can't wait to get on the road again

Paolo, you'll have to make sure you have a copy of that song cued up when you guys command the first long segment.

You're very funny Dan
153m is by far the longest ever, isn't it ?

BTW, nice new avatar. I hope it'll help launching MSL on schedule

I think A and B are right, but not C and D. Check out this post and this post.

Was my "selective Gaussian" method here any good?

Is there anything else I can do to help with this analysis? All of this is amazing!

If I read ypur description of the algorithm correctly the regions, after the edge prserving smoothing, are selected based on brightness (the fuzzy select). I'm not sure that ripples, bedrock and sand are distinguishable by intensity values. The currently proposed filters (variance, texture, fourier transform) try to measure "local differences" so to speak. They are therefore somewhat insensitive to the absolute values of brightness.

Paolo

Many many thanks to all who contributed so far. I want to reiterate something I want to make sure this is constantly clear. When I participate and write here on UMSF I do it as a private citizen who accidentally drives rovers on Mars. The discussion and my request for help was from me, not JPL. Therefore, your contribution will be to me, you cannot claim to have helped JPL drive the rovers or provide a traverse map to JPL. The silver lining is that as a consequence you are not liable for the actual safety of the mission.

This starts to sound like a lawyer speak, doesn't it? Well, it isn't. It comes from me.

I could have attempted to go through the official channels at JPL but decided that I much rather put my limited resources into making this experience as close as possible. This has the disadvantage that you are not talking to JPL but to me. I hope this does not discourage anybody from continuing the discussion and hard work. Just looking at the quality and sheer volume of your contributions makes me really really happy. It has been a good week on good old Earth.

Paolo

OK, thanks. That's good to know.

My final iteration for the Erebus to Etched Terrain section:
Click to view attachment

Differential image shifted 10 pixels to the E AND 5 pixels to S (0.5 wavelength of SW-NE dune) to better light up the scary curvy dunes.

Colors are balanced, too.

Exact Recipe:
1) Load image
2) Select rock areas (using Magic Wand): recolorize as grayscale 160/256
3) Gaussian blur 4 pixels (big dune wavelength = 22 pixels)
4) Copy layer
5) Move upper layer 10 pixels to the E and 5 pixels to the S (big dune wavelength = 22 pixels)
6) Set upper layer combination style as "Differential"
7) Merge the two layers
8) Gaussian blur 30 pixels (big dune wavelength = 22 pixels)
9) Add Curves layer (set 0/256 as black level, white level at 25/256)
10) Add Brightness/contrast layer (set Brightness +75)
12) Flatten layers: convert Image Mode to Indexed Color
13) Color table:
1/256 pixel value = Green[125/100%/100%] (Hue/Saturation/Brightness)
80/256 pixel value = Yellow[65/100%/100%] (Hue/Saturation/Brightness)
160/256 pixel value = Orange [30/100%/100%] (Hue/Saturation/Brightness)
256/256 pixel value = Red[0/100%/100%] (Hue/Saturation/Brigthness)
14) Convert Image Mode to RGB: duplicate layer
15) Reload original image
16) Combine the two images using Normal mode (original image at 30% opacity)
17) Crop slightly to remove edge artifacts

-Mike

I also speak for me only (one that doesn't contribute so far but hope he'll help more on a day by day basis later on) you're statment was clear from the begining.
The idea that "I" can provide you with "choices" even if you'll never choose one (or tell you did) is enough for me to feel I participate in the Exploration.
Since you've started to post the feeling is different. I know you tell us what you're allowed to tell but it's much more than what we had before.
And it's good to know we have an Italian driver but I'm happy Oppy is not a Ferrari; they've not be very lucky at Singapore Grand Prix today...

I like what your doing Mike, but the bigest green area on your map is a known death trap filled with fine dust (the center of Erebus).

Ahhh, but at least it's a smooth fine dust.

I think the shift-differential image technique (and Fourier technique also) will only be able to help avoid areas with dune crests.
There will still be a visual inspection part that will be necessary to winnow the routes down further. But at least there should be significantly less real estate to have to examine closely.

-Mike

The curved dunes usually come in pairs. (There is a curved trough, with two crests).

By shifting 10 pixels to the E and then 5 pixels to the S, it appears that you can get maximal brightness signal from the SW-NE oriented dunes after taking the differential - these are the ones like Purgatory.

These will get lumped into the category of "big, scary" dunes. But it is comforting to know that Purgatory Dune and it's buddies (and there are several in the area) will get indicated as "RED = Caution!"

Here are the comparison images of Purgatory Dune from ElkGroveDan's post (here) and images from the "Differential Shift 10E5S Method". Purgatory and friends indicated by black arrows:
Click to view attachment

(In retrospect, this was kinda scary.)

-Mike

(I don't know if a Fourier analysis would pick this up. It's only a repeating number of 2 instead of a regular set.)