Author Topic: Plate solve . Most accurarate plate solver ?  (Read 3987 times)

Offline Thomas

  • Newcomer
  • Posts: 31
Plate solve . Most accurarate plate solver ?
« on: 2017 June 08 10:31:27 »
Hi,

(sorry a bit Off-Topic)

I experimented with several plate solve programs (via  GUI and also via scripting over VS VB.NET and C++/CLI .NET) and I am curios what would give the very best results to get the center coordinates.

In all my tests with different catalogs (APM, GSC1.1 UCAC3 , USNO-A2.0 . Mostly used  UCAC3 catalog):

PixInsight ImageSolver, PlateSolve2.28, online upload to novas astrometry net -> The difference of the reported center coordinates are within ~ 0.2 ArcSec in RA and DEC

PinpPoint 5.x , PinPoint 6.x and even PinPoint 6 All Sky solving, is mostly ~ 1 - 2 ArcSec off -> Regarding to the results of the above listed plate solve programs.

Off course using different catalogs and changing parameters can affect the result , mostly in a range of ~ 0.2 ArcSec .

I am wondering if anyone else has noticed similar effects and what you are using the get the highest accuracy .


Thomas



Offline Niall Saunders

  • PTeam Member
  • PixInsight Jedi Knight
  • *****
  • Posts: 1456
  • We have cookies? Where ?
Re: Plate solve . Most accurarate plate solver ?
« Reply #1 on: 2017 June 08 16:35:18 »
Hi Thomas,

Thinking about your images returning 'centre accuracies' of less than 1 arcsecond when you plate-solve - wouldn't you agree that this level of accuracy is probably well within the limits imposed on your image data simply due to 'seeing' conditions?

Another thing to consider is the 'resolution' of your images - what sort of 'pixel-scale' are you working with (you would get this from the results of the plate-solve itself)?

You really need to know all of this information if you expect to be able to decide whether one catalogue sourve is better than another, or whether one plate-solve process is better than another. You have to be able to clearly define all of the information that has been used to perform any of the calculations.

This reminds me of when I used to manage and run an electronic calibration laboratory. If we set out to offer a 'Certificate of Calibration' for some device, then that certificate had to show 'traceability' all the way back to some internationally defined 'reference'. Tke for example a sophisticated electronic frequency counter - with an eight-digit display, and which we use to measure a frequency of 1000000Hz. We have  only used 7 of the 8-digit capability of this meter so, let's assume that we can even read down to 0.1Hz. Now our display could return a displayed measurement of 1000000.0Hz.

But, what would it mean if that is exactly what we observed when we made the measurement?

Was our meter 'perfectly' calibrated? Well, that would depend on how accurate our 'source' was (equivalent to the catalogues you are trying to plate-solve against). If the frequency source was wrong, but our meter was also wrong - yet we still see the 'perfect' display, how can we possibly know that we have 'two' errors? Simply put, we can't - it is impossible.

In my case the 'source' itself had to be calibrated - once a year - assuming that it's previous performance did not indicate that even more regular testing was required. Our test conditions also had to be 'perfect' (within tightly controlled limits for temperature and humidity), just to give us the best chance of maintaining a frequency source that was reliable. So, we had to send our 'calibration reference source' away for 'calibration' itself. And the lab that tested it for us had to send their source off for calibration as well - and so on, all the way up the chain to an internationally accepted 'reference standard'. Typically, for frequency sources, this would be a Caesium Beam Frequency Standard, an 'atomic clock, if you will - but, even these have to be 'calibrated' (and, back in the 1980's I was one of only a very small handful of people specially trained to set up such atomic clocks !!).

So, your reference star charts would have to be 'traceable' in as much as the position data they contain for the objects that plate-solving processes work with. And, as I suggested at the rtart of this reply, it is highly doubtful that anyone other than a full-blown specialist photometric observatory would have the required equipment, and specialist knowledge, to be able to call into question the positional data of a ctalogue.

We, as 'normal users' simply have to accept whatever solution is returned to us by whatever process we have used to plate-solve against whatever star catalogue we have selected.

If you have acquired an image that contains some object of special interest, whose position appears to either not to be contained in any catalogue that you have access to, or whose position appears to be significantly different to that contained in the catalougues, then you would be best to submit that image (with whatever positional data that you have calculated) to an authoritative body who can examine your data with higher orders of accuracy.

I hope I haven't discouraged you - but 'measurement accuracy' is a special area of interest to me. I spent my whole electronics career trying to explain this to colleagues (and trainee delegates). I have had seasoned expedition-adventurers, trekking their way 'solo' across the frozen deserts Antartica, contact me over the HF radio, in tears, because they couldn't understand or explain the innacuracies that they were observing between their GPS receivers, their daily log and their magnetic compasses. Some of these seasoned veterans even had to be rescued (from themselves) before they perished, or were otherwise lost forever.

So, please take my comments as positive, and don't try to over-analyse data that may simply 'not be there'. Of course. if I have missed the point in your query, or if others also want to jump in and comment (or even question my thoughts on the subject), I am always happy to see a thread develop and flourish!!
Cheers,
Niall Saunders
Clinterty Observatories
Aberdeen, UK

Altair Astro GSO 10" f/8 Ritchey Chrétien CF OTA on EQ8 mount with homebrew 3D Balance and Pier
Moonfish ED80 APO & Celestron Omni XLT 120
QHY10 CCD & QHY5L-II Colour
9mm TS-OAG and Meade DSI-IIC

Offline Thomas

  • Newcomer
  • Posts: 31
Re: Plate solve . Most accurarate plate solver ?
« Reply #2 on: 2017 June 10 03:20:27 »
Hi Niall,

to make it (hopefully) more clear :

I used several different plate solve programs for a certain picture and their result for the center coordinates are ALL within 0,2 ArcSec . (Ra and Dec)

Then I used Pinpoint5, and 6 and PinPoint AllSky solver with the same picture and the same catalog, and their calculated center coordinates differs 1-2 ArcSec to the other programs .

Of course I did those test with different pictures and different catalogs and the behavior is always the same.

Offline vicent_peris

  • PTeam Member
  • PixInsight Padawan
  • ****
  • Posts: 988
    • http://www.astrofoto.es/
Re: Plate solve . Most accurarate plate solver ?
« Reply #3 on: 2017 June 10 03:38:36 »
Hi Thomas,

Besides this discussion, the ImageSolver in PixInsight has a unique feature: it uses surface splines to correct for geometric distorsions. This is much more accurate than using polynomials or the standard WCS convention. I can show you some examples where it's impossible to correctly solve the astrometry by using polynomials, giving enormous errors at the corners.

Best regards,
Vicent.

Offline Niall Saunders

  • PTeam Member
  • PixInsight Jedi Knight
  • *****
  • Posts: 1456
  • We have cookies? Where ?
Re: Plate solve . Most accurarate plate solver ?
« Reply #4 on: 2017 June 10 03:46:16 »
So, does that suggest to you that the PinPoint method is perhaps not as 'accurate' as the other methods? (I don't know anything about PinPoint, so this is a generalised question).

If you can determine that the PixInsight plate-solve method is at least as accurate as several other methods that you have already tested, then does that not leave you with enough confidence just to rely solely on the PI solver?

Going back to my abstract concept of 'traceability', if you rely on Catalogue XYZ, and use your image, ABC, this eliminates two of the system variables as you try to 'calibrate' your plate-solve process query. The only variable left, now, is the code being run by the process, and the set-up/configuration variables that you provide for the process.

Unfortunately, the selection of configuration data for the plate-solver can be a non-trivial task itself (that's why I hate and detest plate-solving in general - it seems that you, the user, have to do half the work for the process yourself !!). So, that selection can also make it very difficult to now compare process P versus process Q versus process R, etc. You will need to examine the processes that you are interested in, and see if you can eliminate some of the 'variability' between these configuration points.

Then, when you have eliminated all the variability that you can, you need to determine whether the final 'accuracies' of the various processes can be related to, or explained by, those configuration points that have no common ground between the processes.

This is not going to be a simple task - but it is one that you feel you need to complete. Only you can knoow how much effort you are willing to put into making the comparison.

Personally (remember - I hate plate-solving!!), I just want the 'simplest' process - and that means I am happiest working within the confines of PixInsight. At least, by doing this, I know that the plate-solve method would never have made its way into the normal PI release package if it hadn't already passed a significant peer-review by PTeam members. I know that a great deal of effort will have been put into the solver method, and I know that bugs or errors will be identified and eliminated if, or when, they appear. Other solver processes have an element of 'anonymity' that PixInsight just doesn't encourage.

And, that, is good enough for me!
Cheers,
Niall Saunders
Clinterty Observatories
Aberdeen, UK

Altair Astro GSO 10" f/8 Ritchey Chrétien CF OTA on EQ8 mount with homebrew 3D Balance and Pier
Moonfish ED80 APO & Celestron Omni XLT 120
QHY10 CCD & QHY5L-II Colour
9mm TS-OAG and Meade DSI-IIC

Offline vicent_peris

  • PTeam Member
  • PixInsight Padawan
  • ****
  • Posts: 988
    • http://www.astrofoto.es/
Re: Plate solve . Most accurarate plate solver ?
« Reply #5 on: 2017 June 10 05:25:32 »
Let me some time. I'll try to show the difference between distortion correction applied with polynomials and surface splines in some of my images. I'll try to solve one of my telephoto images with an earlier version of ImageSolver (where we used polynomials). You'll see that the difference is huge. I don't mean a sub-pixel difference. I mean an error in the range of 10 - 20 pixels in some cases.

Best regards,
Vicent.

Offline Thomas

  • Newcomer
  • Posts: 31
Re: Plate solve . Most accurarate plate solver ?
« Reply #6 on: 2017 June 11 13:26:18 »
Let me some time. I'll try to show the difference between distortion correction applied with polynomials and surface splines in some of my images. I'll try to solve one of my telephoto images with an earlier version of ImageSolver (where we used polynomials). You'll see that the difference is huge. I don't mean a sub-pixel difference. I mean an error in the range of 10 - 20 pixels in some cases.

Best regards,
Vicent.

Thanks Vicent,

no need to hurry ..  ;D