Forum

Author Topic: accuracy assessments  (Read 6164 times)

drmatt

  • Newbie
  • *
  • Posts: 1
    • View Profile
accuracy assessments
« on: November 11, 2013, 11:09:03 AM »
Hi All.  Here are some recent results of mine using photoscan:   www.drmattnolan.org/photography/2013 

I've been using it about a year now, and couldn't be happier with the simplicity and power of the software.  I'm processing blocks of 1000+ D800 images without issue, and getting camera residuals and repeat-DEM residuals of ~15cm consistently, using no ground control, only accurate camera coordinates.

I'm curious if any one has written or read any scientific papers describing the accuracy of photoscan?  This could be in the form of lidar-photogrammetry differences, or photogrammetry-photogrammetry, etc or a technical paper on the algorithms and their limitations, or a rigorous assessment of the errors inherent with each component of the process (GPS, bundle adjustment, optimization, etc).  I'd also be interested to learn if anyone is getting better than 15cm repeatability or has better ways to assess errors than I have.

-Matt

Limpopo_River

  • Newbie
  • *
  • Posts: 25
    • View Profile
Re: accuracy assessments
« Reply #1 on: November 11, 2013, 08:01:22 PM »
You'll find some examples of comparisons between photogrammetry and LIDAR here:
http://ncptt.nps.gov/blog/category/product-catalog/videos/digital-documentation-2/
This one, in particular, would be of interest:
http://ncptt.nps.gov/blog/a-comparative-study-using-lidar-digital-scanning-and-photogrammetry/

Wishgranter

  • Hero Member
  • *****
  • Posts: 1202
    • View Profile
    • Museum of Historic Buildings
Re: accuracy assessments
« Reply #2 on: November 11, 2013, 09:16:33 PM »
If remember GeeVee had done some interesting test and was published in PDF. its from 2011 if remember correctly.....

And remember article on sockpile precision measurement. so don't remember the actual precise number but:
GPS measurement  3,8 % Diff
LIDAR ( ground ) meass. 2,4 % diff
Photogrammetry meass. 1,9 % diff

was published on SPAR http://www.sparpointgroup.com/ a year or more back
----------------
www.mhb.sk

gEEvEE

  • Jr. Member
  • **
  • Posts: 66
    • View Profile
Re: accuracy assessments
« Reply #3 on: November 11, 2013, 09:47:11 PM »
Hi all,

The geometric accuracy depends on so many parameters, that you cannot give a general statement. I did indeed a lot of comparison and testing (most publications can be found online although many were never published). Generally, both the georeferencing and final positional accuracy can be very high (sub 10 centimeter) but this will generally not be achieved by only pressing the three workflow buttons. You need to know how to take the images, how to optimize the SfM results, which GCPs to take, which images should contribute to the final mesh and ortho, which camera parameters to optimize etc. So, only playing with PhotoScan and the experience that comes with it will tell you.

Also, sometimes it is not all about geometric accuracy: from time to time, the colour info can be more important that the absolute accuracy. So even in those situations that the SfM+MVS results are less accurate than ALS (Airborne LaserScanning), an image-based approach has its own pros (while also ALS can have its cons). For instance, most people are not aware that ALS data are often not properly georeferenced. No individual strip adjustment is done, resulting in small but sometimes nasty georeferencing errors leading to all kinds of artefacts. In those cases, SfM+MVS might even have the edge.
Those of you that want to know more about ALS, SfM+MVS and related stuff, feel free to contact us or check the publications of our Archaeological institute: http://archpro.lbg.ac.at/publications/publications-lbi-archpro. Most of the publication that I (co)authored can be found on my academia website.

Best regards,
Geert

skkiester

  • Newbie
  • *
  • Posts: 3
    • View Profile
Re: accuracy assessments
« Reply #4 on: February 14, 2014, 10:24:56 PM »
Hey drmatt,
I was wondering if you were getting 15cm accuracy with just camera positions or if you had incorporated orientation as well? I think it would be nearly impossible to get that kind of accuracy with just position, but I wanted to make sure.

I am an undergraduate student at the University of Alaska in Anchorage and I have done some research on SFM accuracy (and will be continuing to do so). As for now, I think gEEvEE's statement is absolutely correct. Accuracy for PS is largely dependent on many factors, not the least of which are focal/baseline length and point density (resolution). I can say that with appropriate ground control I have always achieved an acceptable accuracy for volume calculations. We did a comparison of SFM vs. a terrestrial laser scanner and ended up with about half the accuracy with SFM as we achieved with the scanner (based solely on control points located around the site).  One definitely has to use independent checks as well, because the accuracys given within PS are not indicative of the actual accuracy.

Here is one technical note on the subject:

http://www.blm.gov/nstc/library/pdf/TN428.pdf

Cheers,
Seth