Forum

Author Topic: Photoscan Model accuracy  (Read 6691 times)

Areslow

  • Newbie
  • *
  • Posts: 3
    • View Profile
Photoscan Model accuracy
« on: April 25, 2016, 01:05:35 PM »
Hi guys
where can i find the model accuracy in photoscan ? i want to know the standard deviation of points height in  my model , is this info available some where in photosca  ? actually i want to know the vertical and/or horizontal accuracy of my model .

frank.stremke

  • Full Member
  • ***
  • Posts: 206
    • View Profile
Re: Photoscan Model accuracy
« Reply #1 on: April 25, 2016, 06:04:01 PM »
hi
well there are the error values in the control tab but they are of course only of limmeted use to understand the models accuracy since they are not representative of the final result and in most cases not independend
i have developed my own routine to create decent statistics on the accuracy
since iam in most cases interested in vertical accuray this is simple and can be done mostly automated
i use the DEM and an additional dataset of ground check points the more the better. Then i run a coparison of booth values in QGIS and can have the result as a graph or a table including standard deviation and 2xSD as well as all other kind of data i like
my last big model had 31000 check points and with this data you can do some serious statistics
frank

Airgo

  • Newbie
  • *
  • Posts: 35
    • View Profile
Re: Photoscan Model accuracy
« Reply #2 on: May 13, 2016, 02:03:16 PM »
Hi Frank,
can you tell us what average vertical accuracy you get?
Mathias

stihl

  • Sr. Member
  • ****
  • Posts: 410
    • View Profile
Re: Photoscan Model accuracy
« Reply #3 on: May 13, 2016, 02:28:24 PM »
The vertical accuracy depends entirely on your project GSD, individual image quality and quality of the entire dataset.
The lower the GSD (ie the images have a high zoomfactor or the flying altitude is low) the more accurate the vertical accuracy is in theory.

Generally speaking the proposed accuracy is 1-2 times the GSD for the X and Y and 2-3 times the GSD for Z.
For instance if you're flying a project with a native GSD of 5 cm, your expected Z accuracy would be between 10 and 15 cm in ideal conditions (ie sharp and well lit images, high overlap (80/60% for instance))

Obtaining an aboslute vertical accuracy less than 3 cm with RTK GPS is achievable with enough GCP's which need to be well distributed in the project area.

Airgo

  • Newbie
  • *
  • Posts: 35
    • View Profile
Re: Photoscan Model accuracy
« Reply #4 on: May 13, 2016, 03:23:37 PM »
Thank you!
That means if my images have a GSD of 1,5cm the z accuracy will be less the 5cm?
In our UAV there is no RTK-GPS. I assume you mean that the GPS for the GCP's should be a RTK-GPS to achieve the accuracy of less than 3cm, right?
What does "well distributed GCP's" mean? What should be the minimum distance between them? How much GPS's are required tor an area of 120 hectares?
That you very much for your help!

stihl

  • Sr. Member
  • ****
  • Posts: 410
    • View Profile
Re: Photoscan Model accuracy
« Reply #5 on: May 13, 2016, 03:51:27 PM »
How many and where they are needed for a terrain of 120 hectares is hard to say.
It's wise to place GCP's on at least each corner of the terrain and within the project area they should be well distributed.
From trial and error I can say that a radius of 150m around a GCP is sufficient to provide decent coverage. More is better, but it's obviously more work in the field.
From my own experience I can say that If you have RTK GPS in your UAV or airplane then you can reduce the needed amount of GCP's by up to 90%.
The GCP's should always be measured with RTK GPS to obtain a sub 5cm vertical accuracy. With a GSD of 1.5cm and enough well measured GCP's you can achieve such accuracy.

Airgo

  • Newbie
  • *
  • Posts: 35
    • View Profile
Re: Photoscan Model accuracy
« Reply #6 on: May 13, 2016, 04:26:31 PM »
Thank you.
Just to be sure:
If the GCP's are enough und well measured with  RTK-GPS I do not need a RTK-GPS in the UAV?
But if I have a RTK-GPS in the UAV I can reduce the amount of GCP's?

stihl

  • Sr. Member
  • ****
  • Posts: 410
    • View Profile
Re: Photoscan Model accuracy
« Reply #7 on: May 13, 2016, 06:00:26 PM »
Correct to both statements.

Dave Martin

  • Full Member
  • ***
  • Posts: 170
    • View Profile
Re: Photoscan Model accuracy
« Reply #8 on: May 14, 2016, 11:27:58 AM »
Thank you.
Just to be sure:
If the GCP's are enough und well measured with  RTK-GPS I do not need a RTK-GPS in the UAV?
But if I have a RTK-GPS in the UAV I can reduce the amount of GCP's?

Airgo,

just to add to Stihl's comments, the better the GPS fitted to the UAV, the better the results, but it need not be RTK - so a WAAS-augmented receiver is better than one without; a Glonass capable one is better than one without, and one that can process Gallileo and/or Beidou etc. is even better. Also, some promising results are being achieved with post-processing a suitable RINEX-logging carrier-phase receiver (so not RTK / no live correction link necessary but you do need access to high-sample-rate base station data to carry out the post-processing).

GCPs can be most easily measured with RTK GPS but small numbers can be measured by static GPS occupation; and can also be measured / "densified" using total station etc.

Dave