Forum

Author Topic: Alignment not good  (Read 8612 times)

gheflorian

  • Jr. Member
  • **
  • Posts: 50
    • View Profile
Alignment not good
« on: May 25, 2013, 05:01:16 PM »
https://docs.google.com/file/d/0B4FwEb6tk6g7cUVpZHRLOTdMS28/edit?usp=sharing

I have tried everything from sharp to smooth alignment. I have entered 50 on hole threshold and put mild on depth filtering but still the result is not good. Can anyone help me?
Alignment was done using high accuracy. Thank you!

Wishgranter

  • Hero Member
  • *****
  • Posts: 1202
    • View Profile
    • Museum of Historic Buildings
Re: Alignment not good
« Reply #1 on: May 25, 2013, 06:02:17 PM »
how many camera positions ? overlap ?? try put here screen with cameras...... and some info from CHUNK....
----------------
www.mhb.sk

gheflorian

  • Jr. Member
  • **
  • Posts: 50
    • View Profile
Re: Alignment not good
« Reply #2 on: May 25, 2013, 06:55:45 PM »
https://docs.google.com/file/d/0B4FwEb6tk6g7RTVic056T01xSW8/edit?usp=sharing
1255 photos. Overlap from the flight (UAV) is 75 with 35. Flight altitude 150m.

gEEvEE

  • Jr. Member
  • **
  • Posts: 66
    • View Profile
Re: Alignment not good
« Reply #3 on: May 27, 2013, 06:59:39 PM »
I see that you have GPS coordinates stored in your pics. Try to load your pics in a new project and uncheck all the pics with GPS coordinates in the ground control pane (so that they are no longer taken into account).

Verify that the focal length and 35 mm focal length are embedded in the Exif data (if not, fill in all parameters in the camera calibration dialogue or embed this info into your Exif and load the pics again)

Afterwards, set the amount if interest points to about 25 000 and align on high accuracy with pairs enabled (certainly for UAV flights, pairs should always be enabled as there is normally a very systematic data acquisition).

Now align the data and check for badly aligned images in a visual manner. If you find them, reset the camera alignment and try to align this/these image(s) again. Afterwards, filter on reprojection error. If the data set is well aligned, you should be able to filter all interest points with a reprojection error above 1 pixel without deleting more than 1/5 of your complete sparse point cloud. Now optimize with everything checked apart from aspect and skew.

This should give you a good alignment. Verify this by building a low quality model. If you use GCPs, build the low quality model without optimization. Afterwards, indicate the GCPs and optimize as described. Before you optimize with GCPs, make sure you set the coordinate reference system and accuracies of GCPs in the ground control properties.

Cu,

Geert