Curiosity Image Retrieval Tools, scripts and software Options

[jmknapp]

Be warned that the current MAHLI depth maps are not calibrated for distance...

That was going to be my next question.

For RSM instrument boresighting, I find it hard to believe that the offsets you're seeing are purely parallax (as noted they are in the wrong direction) but this is something I'm just starting work on so I could be mistaken.

We will all enjoy that work very much, Mike!

[Gerald]

Thanks for this info! So we have to look for a way to do a depth calibration. This won't be quite trivial.
There might be a way to reconstruct/guess the focussing heuristics; e.g. a hyperbolic interpolation of the type D = C + A / N, for N = N_min..N_max, and some constant distances A and C.
If the heuristics cannot be retrieved in a straightforward way, we'll need a somewhat fuzzy optical technique.
One idea for that might be the simulation of the optical blur of single (lo-res) images from the sharp focus-stacked image. Via difference image and a chi-square method (minimizing the sum of the squares of the difference pixel values of the examined region) and constraints known of the focussing heuristics we could try to find out the most likely focus distances.
That's just a first idea and has to be elaborated in more detail, if necessary.

[mcaplinger]

If the heuristics cannot be retrieved in a straightforward way...

My suggestion is to wait until the PDS archive data are available so you don't have to guess. We have enough trouble figuring out the range as it is. By that time it may be that the JPL stereo mesh products will be available (I don't know if those get archived to PDS or not); if so, you will know the range out to 20 meters or so for each site. I'm also not sure what you have to gain; usually you can tell where MAHLI is looking just from looking at where MAHLI is in Navcam images, and you can tell where MAHLI is in the rover frame from the SPICE data.

If people still want to look at this, the MAHLI paper has some focus/range plots and all Z-stacks are taken with constant motor count stepping (but I'm still not sure if absolute range can be backed out, at least without some measure of scale.)

[Gerald]

Thanks, that's perfect! All infos we need.
And thanks for your hard and excellent work!

EDIT: For technical data see Curiosity’s Mars Hand Lens Imager (MAHLI) Investigation, especially section 6.5 on p. 25f, and Fig. 27 on p. 102 of the submitted version.

[Greenish]

Won't be too long now, according to http://pds-imaging.jpl.nasa.gov/ , "The first MSL release for EDRs is expected on February 27, 2013; the first release for RDRs is expected on March, 20, 2013."

[Deimos]

Again, red is the automatically generated position and green the actual. So in this case it's off more to the left. Spherical geometry makes my head hurt too, but it's hard think of why the direction would be so variable--it's not like the rover is tilted so much. Of course, the calibration target is very close, so any parallax is that much greater.

I grabbed those images and ran my own models. I looked at where the center of the CCAM FOV was in the NCAM image for a set of assumed distances (every 10 cm to 100 m). Your red circle is near where I see the far end of that projection. The projection goes through your red circle just shy of 1.6 m. Repeating for the NCAM right eye, it also goes through that target shy of 1.6 m (from CCAM entrance aperture).

In the left eye, the projection enter from 1/3 down the right hand side of the image. In the right eye, the projection enters about 1/4 up the left hand side of the image, with the far end firmly in the RTG fins.

So the effect seems to be purely parallax from the vertical + horizontal offset of the cameras, with the left eye having a stronger vertical component.

[jmknapp]

Good to know! Thanks, deimos.

[Gerald]

Explanations to this Sol 230 MAHLI image contain the following data points of a motor count - cm-distance - map for MAHLI with opened dust cover:

CODE

2.5;15270
5;14360
7;13980
10;13635
15;13325
20;13155
25;13050
30;12970

By a least squares method I got this formula for a regression parabola describing the working distance in cm by motor count:

CODE

distance = 13279.7 / (motor_count - 12563.9) - 2.40117

applicable for motor counts between 12564 and 15600, with care outside the 2.5 to 30 cm region, because no data points were avaiable there.
Absolute and relative errors of the formula relative to the data points are shown in these diagrams:

[jmknapp]

Wow, that was a geeky-informative note they added to that image. Which motor does the count refer to?

[mcaplinger]

Which motor does the count refer to?

There's only one motor in MAHLI, the focus motor.

Working distance to motor count is described in the MAHLI paper, figure 13.

[Gerald]

It refers to the "focus motor count position" of the MAHLI camera. The distance may be useful for a 3d localization of objects we thought about earlier.
There are also provided micrometers per pixel, but less accurate. I'll see, whether I can retrieve an acceptable formula, that interpolates this second value. It may be interesting for a scaling of the MAHLI images.

All we need is the appropriate motor count. Joe, do you know how to access to it?

[Gerald]

Slightly improved hyperbola formula for MAHLI distance, and formula for microns per pixel, based on same explanation:

CODE

distance = 13407 / (motor_count - 12558.2) - 2.44264
MicronPerPixel = 46836.6 / (motor_count - 12560.1) - 1.20623

Errors relative to data points below 1%:

[jmknapp]

Like most metadata not in the PDS, I wouldn't know how to get the MAHLI information, but maybe it will be included in the MAHLI PDS release scheduled for June 6 ("release 2").

[Gerald]

This is a link to an arbitrary MAHLI image: http://mars.jpl.nasa.gov/msl/multimedia/ra..._DXXX&s=129.
The adress is composed of the fixed string "http://mars.jpl.nasa.gov/msl/multimedia/raw/?rawid=", the name of the jpg-file without extension, here "0129MH0154002000E1_DXXX", the constant string "&s=" and the sol number, here "129".
The referenced website containes the constant text "When this image was obtained, the focus motor count position was " plus the motor count (here "15148") plus ".". There are no html tags in that part of the html source.
By scanning (searching) for the fixed text, then reading the number until the "." it should be possible to read out the motor count.
What do you think?

[jmknapp]

Huh, didn't know they had that for every MAHLI image--seems like that info could easily be scraped and converted to cm as you say. I may try that for my website. Just did a quick check & there have been 4224 MAHLI images to date, so that's a fair bit of scrapage.