Agisoft Metashape
Agisoft Metashape => General => Topic started by: radar_boy on December 20, 2016, 06:31:22 PM
-
I'm using some vertical aerial photographs to create a Digital Elevation Model (DEM) and orthophoto of the terrain. I have ground control points (GCP) measured using Differential GNSS. The images align very well and the production of the dense point cloud (setting High) goes according to plan. However, in the point cloud there are gaps, where patches of ground are noticeably higher or lower than that surrounding it. It is hard to see without panning around, but when creating a mesh from these points the patches and their edges become really evident. This is nothing like the actual landscape. How are these caused and how can they be corrected?
Many thanks
(https://c5.staticflickr.com/1/536/31618468612_f06e8f2b2f_o.jpg)
-
Hello radar_boy,
Such artifacts usually indicate the issues with the camera alignment.
Can you provide the information about alignment/optimization settings used and post a screenshot with residuals page from the report file?
-
I used Medium for initial alignment.
I then used Gradual Selection and culled the points according to the following threshold:
Projection Error: 0.2
Reconstruction Uncertainty: 20
Projection Uncertainty: 20
I then Optimised Cameras with defaults using the remaining points.
-
Here is the report:
# PhotoID, X, Y, Z, Omega, Phi, Kappa, r11, r12, r13, r21, r22, r23, r31, r32, r33
os_82033_v_007.tif 385919.2062696178900000 80746.1490088077700000 2427.9801713772022000 0.7060472463219131 0.2065422957097703 107.2342390549936500 -0.2962769314526411 0.9550158037652961 0.0128372289662858 -0.9550952753791041 -0.2962987877252764 -0.0002081851235714 0.0036048352973428 -0.0123224571842452 0.9999175777089938
os_82033_v_008.tif 386584.1999433031300000 80927.3854616020980000 2425.6020255756343000 -0.5128383450749853 -0.4169149396235561 105.5489794258962300 -0.2680549433469141 0.9633456018838577 -0.0105734894103801 -0.9633761468246594 -0.2681140473115296 -0.0046105708435342 -0.0072764741843152 0.0089503611804028 0.9999334697659568
os_82047_v_412.tif 386081.3126213328800000 81161.9456559692480000 2431.9811098230434000 -0.2335982434049884 -0.6860605727900033 91.5673559836439070 -0.0273501499821671 0.9996162176851944 -0.0044030258572207 -0.9995542001773864 -0.0274006826591848 -0.0118576345683536 -0.0119737297321647 0.0040767549051963 0.9999200017330107
os_82047_v_413.tif 386693.7980145422600000 81184.8961358359810000 2436.3725434047310000 -0.0936927697184504 -0.1114061128241012 92.0538592661327900 -0.0358388622876644 0.9993561290843859 -0.0017038812747297 -0.9993556900572562 -0.0358420596531271 -0.0018845450398761 -0.0019444022504265 0.0016352434969241 0.9999967726340895
-
Hello radar_boy,
With such threshold for Reprojection error it may happen that you remove a lot of valid tie points, leaving very low number of tie points in the cloud.
Low number of matching points between images may lead to skipping such pairs by the depth maps calculation process. You can check the number of valid tie points between of each image pair in the View Matches dialog accessible from the Photos pane context menu.
-
With those settings I get 6000+ tie points remaining, from an initial 11000+. When viewing the points on the photos they are also well distributed.
I've tried creating a dense point cloud without culling any points (i.e. the full 11000+) and I get the same patch features.
-
BTW, before any culling of tie points none of them have a Reprojection Error exceeding 0.7
-
Hello radar_boy,
Can you send a report from this project (File Menu -> Generate Report option) to support@agisoft.com? Or even, if possible, the project file re-saved in PSZ format (without depth maps, dense cloud or mesh - just alignment results and control points).
-
Hello radar_boy,
Thank you for sending the project file.
I can suggest to optimize the camera alignment with all parameters checked on, except K4, P3 and P4, although it will discard the dense cloud.
For similar small subsets of images, I can suggest to disable the preselection, use High accuracy and disable Adaptive camera model fitting option in the Align Photos dialog. After the alignment and placing the markers, it's recommended to optimize the alignment, as stated above.
-
I'm revisiting this thread because, unfortunately, I was never able to get rid of these artifacts. I'd hoped that new versions might provide a fix (I'm now running 1.4.1) but to no avail. I followed the instructions in the previous post to the letter, but with no improvement.
-
Hello radar_boy,
Can you please send a new report file from the version 1.4.1?
-
Here it is: http://www.mediafire.com/file/lb3luqhv8gf54rb/Report%20for%20OS%201982%20vertical%20aerial%20photographs%20March%202018.pdf
-
Hello radar_boy,
Can you please try the following:
- export the markers to XML file,
- remove markers from the project,
- in the Camera Calibration dialog clear all values in the Initial tab, except F,
- re-align the images using the same parameters that you have used,
- import back the XML file,
- optimize camera alignment (using the same parameters you've used before). Before that you can also modify the "marker accuracy (m)" parameter to lower accuracy, if it is set to 5mm at the moment, for example, to 10 cm.
-
I can't export as XML, when when I import the same XML nothing appears (there are no error messages either)...
-
Hello radar_boy,
I meant to use File Menu -> Export Markers and File Menu -> Import Markers.