Forum

Author Topic: Problem aligning images and gcps  (Read 1878 times)

Guy Byrne

  • Newbie
  • *
  • Posts: 2
    • View Profile
Problem aligning images and gcps
« on: August 06, 2020, 07:23:12 AM »
Dear Forum,

I'm still on trainer wheels with metashape (v1.6) but after dozens of iterations to troubleshoot a problem ... i think i need some advice.

I have faithfully followed the  Agisoft 'Tutorial for beginners: Orthophoto and DEM generation (with GCPs) ..' and a number of (excellent) video tutorials by SkySurv.


My test data set comprises imagery captured with a dji mavic pro ... and a web of rtk GCP,s captured using an altus aps3g base/rover gps.  The photography have 80% forward and side overlap, is I assume in wgs84 .  Flying height was ~30m above ground on a perfect day with little or no breeze.  The GCP's are in GDA/94  UTMS (zone 55) (SE Australia).


Briefly i can get a very good result if i use a small subset of my photos (76) and the GCPs (6) - but when I try and emulate the workflow with my whole data set (280 photos and 15 gcp's)  I can't get a usable solution. 

The first manifestation things are array is after i align the 280 photos ..  there is an east west offset (about 27m) in the 'cameras' and the sparse point cloud.   When I bring in the GCP's that offset persists.

I have tried both transforming the imagery to GDA and also leaving it in wgs84, i have also tried explicitly defining the different image and gcp coordinate spaces.

I can import my gcps as a kml into google earth and they align perfectly .. so i think the error is coming from the photography.
 
I have attached 4 images that i hope explain the problem. 1) a nadir view of the basic camera model, 2)  screen view of the gcp import window, 3) ortho view of the gcp markers and the cameras (is a sensible relationship), 4 ) the camera's and gcp's atop the cloud with the 27m offset shown

thanks in advance

Guy Byrne
(Geoscience Australia)

« Last Edit: August 06, 2020, 07:24:45 AM by Guy Byrne »

Paulo

  • Hero Member
  • *****
  • Posts: 1303
    • View Profile
Re: Problem aligning images and gcps
« Reply #1 on: August 06, 2020, 07:56:37 PM »
hI gUY,

Maybe you can show a screen capture of the camera calibration adjusted values tab to see if there are unrealistic values for f and cx, cy....

I have seen something similar in following post with Mavic Pro 2 imagery..https://www.agisoft.com/forum/index.php?topic=12178.0

As shown in attached screen copy, a normal alignment of this Mavic Pro 2 data set can cause the point cloud being shifted relative to camera positions... This is due to unreliable and huge adjusted principal point coordinates (cx and cy). This happens when camera is always pointing in same direction regardless of flight line. For example for EW flight, camera top always pointing east...as seen in estimated Yaw being around 90 degrees for all fotos regardless of flight line:
Code: [Select]
Foto Yaw (est) Pitch (est) Roll (est)
DJI_0001.JPG 90.243353366879418 0.13425668055079837 1.0882178873775381
DJI_0002.JPG 90.209560282848983 0.15104100995038078 0.80053691006556882
DJI_0003.JPG 90.194902208377655 0.0057662594608951833 0.60124371706134394
DJI_0004.JPG 90.224836087788248 -0.14546689409588004 0.39593194597468195
DJI_0005.JPG 90.310252890189588 -0.38211494371564358 0.20390456465618592
DJI_0006.JPG 90.337949557864391 -0.65572524438963553 0.15180811569945799
DJI_0007.JPG 90.055028766889137 -0.70959502734978053 0.09936390841514621
DJI_0008.JPG 89.906171629394777 -0.6030683284314714 0.22977023337346422
DJI_0009.JPG 89.813317400149771 -0.38555078224892342 0.39428176259088493
DJI_0010.JPG 89.737088502064793 -0.19753743473606639 0.53654718833547965
DJI_0011.JPG 89.709325664923014 -0.024150409593958654 0.65576514107942419
DJI_0012.JPG 89.716553097394225 0.14689542928117783 0.74442989122814252
DJI_0013.JPG 89.714204410905879 0.31990316334798963 0.78620495263656731
DJI_0014.JPG 89.743496456364525 0.59437938450113859 0.85292141678597388
DJI_0015.JPG 89.794402778367157 0.59316424578032279 0.72733746794130194
DJI_0016.JPG 89.920586795661194 0.75922738020845137 0.72584161185672225
DJI_0017.JPG 90.097461346666478 0.58869646187828972 0.57553410532592697
DJI_0018.JPG 90.180194698151311 0.34622632467772152 0.50779815776406578
DJI_0019.JPG 90.265949045763506 0.11976771027785361 0.39721739472769235
DJI_0020.JPG 90.343000936935567 -0.084662045555975785 0.28548993298990394
DJI_0021.JPG 90.43590510079845 -0.28741523127218405 0.11633130145769435
DJI_0022.JPG 90.49243610060519 -0.4776741438752945 -0.0536700695747983
DJI_0023.JPG 90.564167135564531 -0.77753281425227216 -0.18632730813835391
DJI_0024.JPG 90.5765126078565 -0.84919420177709926 -0.20744471524296365
DJI_0025.JPG 90.266461999716924 -0.8316087203334156 -0.15630555668733942
DJI_0026.JPG 90.077318740138111 -0.67916633196636589 -0.012628336675861662
DJI_0027.JPG 89.899038210262162 -0.42783419592236605 0.11480936740591635
DJI_0028.JPG 89.767779923234414 -0.25033720478273813 0.27225790624907786
DJI_0029.JPG 89.694710051934038 -0.098601910960339101 0.38058848259855244
DJI_0030.JPG 89.640569669485402 0.075846329468686829 0.48111444053135582
DJI_0031.JPG 89.640126083003963 0.24338155738064451 0.57526204275408921
DJI_0032.JPG 89.678704339132423 0.50534282521944196 0.61761091456939377
DJI_0033.JPG 89.721332455460953 0.50595934453655367 0.47187786462362752
DJI_0034.JPG 89.854741645428348 0.60493897305137057 0.40528477572087163
DJI_0035.JPG 90.03027318833206 0.4739922775978615 0.17284240786717403
DJI_0036.JPG 90.12108390832465 0.27034017600740345 0.090563180666181464
DJI_0037.JPG 90.203327654297553 0.13836916371221514 -0.022226078576920761
DJI_0038.JPG 90.234199028207087 -0.05193222683361523 -0.1361401480095088


To avoid this, I recommend doing alignment  using both Adaptive camera model fitting and Enable Rolling Shutter compensation.
« Last Edit: August 07, 2020, 04:49:28 AM by Paulo »
Best Regards,
Paul Pelletier,
Surveyor

Guy Byrne

  • Newbie
  • *
  • Posts: 2
    • View Profile
Re: Problem aligning images and gcps
« Reply #2 on: August 07, 2020, 04:01:10 AM »
Thank you Paul!

That thread : Point cloud / Ortho are skewed/tilted (no GCPs)  gives me a lot to work with ..  I'll post an update later.

kind regards

Guy