Support Website Contact Support Blog

Evalution - Skewed Calibration locations

I’m currently using 4.4.9 to process images from a mavic 2 pro.  While I realize it’s still in beta, I have the GPS position issue reported by others (altitude etc).  For this reason, I’m running the beta.  With that out of the way, I’ve flown some test flights today and regardless of automatic of a few manually placed tie points I’m getting a consistent planar skew to my camera position locations (see image).

 

The output from the initial processing step is as follows,

 

 

I am running a free flight at the moment but manually flying a solid overlapping grid.  I had similar issues with another test set and was able to coax it into working, but I keep running into this similar issue.  I’m wondering if it’s something I’m doing wrong or if it’s just something with the beta and my configuration.  

As a follow up, this came off the cloud version.  It’s seeming correct, but the terrain is nearly dead flat as this has it sloping downward.

 

Hi Martin,

I have processed your data set and I obtained much better results:



Here are described the processing options for step 1 I used:

  • Set the image scale on 1/2
  • Free Flight or Terrestrial
  • Use Geometrically Verified Matching
  • Set the internal Parameters Optimization on “All prior”
  • Edit the shutter model of the camera on “Linear Rolling Shutter”

Hope this will help you to improve the quality of your project,

Best, 

Thank you for the pointer.  I ended up using the Rolling Shutter compensation and it did fix the issue.  On a larger run, I’m seeing a similar issue, but only on one end of the computation.  This area should be relatively flat.

 

as evidenced by the overall calculation,

 

I can’t upload this one as it’s large and took roughly 28 hours to compute the Tie points (1438 20Mpix images).  I’m going to reattempt but I’m wondering if there is any way to speed the computation and just use the geolocation points with a heavier bias.  I believe they are much closer to reality than the matched and computed values.  This area is roughly 3 miles of a canal that should be gently sloping from left to right.  I had excellent GPS (16-19 satellites) on a Mavic 2 Pro.

 

Update: I did attempt to upload and got blocked by the “same name” issue.  I’ll append a date to them and try again from work.

I reran with a larger set of data.  This time the skew occurs at the other end.  Everything else looks reasonable, but for some reason the “computed” camera locations just go haywire.  Please note that I’m using 4.4.9 due to a Mavic 2 Pro.  I’m going to try and upload a subset to the cloud compute section so that experts can take a look.

Seam between two segments,

 

I reprocessed with slightly different settings under the “Custom” section under matching.  The results are much better without the significant skew, but I’m still having some mismatch on the flight end alignment (see last image above). 

Hi Scott,

In relation to the large dataset that you are processing - the corridor projects are more prone to inaccuracies. For those projects, the accurate geolocation of the images, as well as the efficient overlap, is very important. It prevents from many issues such as multiple blocks or big errors on the edges that also occurs in your project. If there is a weak area  (e.g. low overlap, low image content etc.), the error will increase as diverging from the breakpoint. Sometimes even GCPs may not manage to adjust the project correctly. 

Your project consists of 872 images. Twenty-eight hours to compute only the Tie points is way too much. Could you share with us your computer specification - CPU, RAM, GPU so we could have a closer look at your hardware attempting to speed up the processing. Geolocation of the images always makes the process go faster, but from my understanding, you used already that information, am I right?

At this point, I need to ask for a workflow explanation. I understand that you used your Mavic 2 Pro having very good GPS signal during acquisition. However, I have a lack of information on:

  • How many flights did you have? Three? 
  • Did you decided to process all the images together or you split the dataset per subprojects and merged them together?
  • Does the project on the cloud contain already your custom processing options?
  • Did you acquire any GCPs for this particular project? I understand that in such a wild area it’s very hard to collect some GCPs. However, this information would significantly help us with this project. 

I look forward to your response so I could give you some processing advices. 

Best!