Agisoft issues

I have a 40-acre field that I would like to map every week using the same flight path. The flight last week went great and Agisoft stitched and processed all pictures with great results (around 250 pictures). Yesterday I do the exact same flight with the same number of pictures, and Agisoft can barely stitch half of them.

For last week's flight it reported an average altitude of 110 m (it was actually 100 m). But for the pictures that Agisoft was able to stitch it claims that the altitude is 80 m when it was really again 100 m.

What went wrong? I didn't do anything differently. Like I said, same flight path, same camera settings, same processing options. Any ideas are welcome.

You need to be a member of diydrones to add comments!

Join diydrones

Email me when people reply –


  • I would tighten the flight path. Or fly 2 laps of the flight path. I combined the pictures from 3 flights and got good results.To get more overlap I was thinking of using 2 cameras mounted on the ends of a 6 ft boom. The cameras don't have to be synchronized. It's just an idea but I think it might require identical cameras with the same settings.3702173074?profile=original
    The domain name '' is parked at Register.TO Domain Registrar
  • Off topic but what benefit is it to collect such data weekly? What are you observing.?

    • Plant growth.

  • Hi Tom,

    I do similar collections with an IRIS+ and GoPro combination and have not seen Photoscan make this type of error.  The errors you quote on the altitude are very large (10-20m).  With my IRIS+/GoPro I generally see errors of less than 3m after initial alignment. Gerard had some good suggestions.  A couple more I might offer:

    1.  What type of camera are you using? If it has a fisheye lens (GoPro or DJI Phantom 2 Vision or Vision+) then make sure you have fisheye selected under the Camera Calibration settings.  Given your large estimated errors it may be that the calculated calibration coefficients have gone all wonky on you.   This can happen if you have a fisheye lens camera and are processing as a frame camera.  

    2.  What version of Photoscan are you using?  I spoke recently with the developers and they said that in the newest version they are now using the camera positions a bit differently during alignment.  If the positions are actually off as much as you say, then they may not be aligning because of this.   Under the reference settings you can change the expected camera accuracy.  Not sure this will fix it, but worth a try.  

    3.  Based on Gerard's first suggestion is to visually verify that you have sufficient overlap and sidelap for processing.  You should be able to do this pretty easily just by scanning through the images. It may save you from chasing other things if for some reason you have gaps that Photoscan can't process across.  

    Hope it works out for you.  I am troubled by the altitude errors.  Those are pretty large.  


    • Not sure why I called you Tom.  Sorry Eric.  

  • Some things to check:

    - was the camera zoomed in? That could reduce your overlap.

    - what is the quality of the images, are they heavily blurred?

    - not sure if agisoft relies on having GPS positions in the exif to do some kind of initial matching. Check if the positions are there (could be a step that was forgotten too). 

    - what is the exposure for the pictures, are they all about the same?  Not all feature matching algorithms can deal with differences in exposure. Could have an impact too because of cloud cover conditions.

    - if the gps positions in the pictures are completely wrong, then perhaps the radius it considers to match with other photos is off. Then you could try another initial matching method (or match every picture against the other) to see if that improves results.

    • Thanks for the suggestions.

This reply was deleted.


DIY Robocars via Twitter
14 hours ago
DIY Drones via Twitter
RT @chr1sa: My talk on PX4 and FAA certification is coming up at 1:45 PST today on the PX4 Dev Summit livestream. Includes some cool new st…
DIY Drones via Twitter
RT @seesharp: I'm tuned into the PX4 / Dronecode free live conference. Great stuff. Microsoft AirSim talk in 10 minutes.…
Jul 6
DIY Robocars via Twitter
RT @masato_ka: 距離センサを3つとESP32を付けたラジコンカーをDonkeyCarライクにNNで自動走行。3層FC極小モデルをTensorFlow Lite for microcontrollerで動かしてる。機体は借り物でRumiCarって言います。Tenso…
Jul 5
DIY Robocars via Twitter
RT @SmallpixelCar: My car was able to go all the way autonomously until the crosswalk. It was only 100 yards from the target. What should b…
Jul 4
Liam left a comment on Agricultural UAVs
I'm Liam from T-MOTOR. I would like to reach out to see if there is any possibility for us to work together.
We are a propulsion system manufacturer who offers motors, propellers and ESCs for all kinds of drone applications which vary from secur…"
Jun 30
DIY Robocars via Twitter
RT @SmallpixelCar: Smart move. The car used the shadow to guide it through the bridge. This was never in the training samples. But it learn…
Jun 30
DIY Robocars via Twitter
RT @SmallpixelCar: Getting closer to the target. Single camera. Untrained road.
Jun 29
Richard Cox left a comment on Australia
"Anyone in the DIYDRONES Australian subgroup based in Alice Springs, NT?
I am experimenting with Ardupilot (standard Arduplane), Pixhawk 4 FC in a 4-ch
RC "AXN Floater Jet" foamy plane..."
Jun 29
Omar Sykes left a comment on Australia
"Hi everyone, I am looking for someone who is good at drone building, repair and software in Adelaide. Please give me a call on 0477 319 219."
Jun 29
DIY Robocars via Twitter
RT @RoboticMasters: #donkeycar
Jun 29
DIY Robocars via Twitter
Jun 29
DIY Robocars via Twitter
RT @RoboticMasters: Donkey car, car car car car car car; Donkey car, car car car car car car; Donkey Car. Anyone like our tiny tiny donkey…
Jun 29
DIY Robocars via Twitter
RT @SmallpixelCar: After improving DBSCAN speed, I can get 11 frame per second on @NVIDIAEmbedded Jerson Xavier MAXN mode and the autonomou…
Jun 26
DIY Robocars via Twitter
RT @Heavy02011: Join us at next Virtual Race League: ⁦@diyrobocars⁩ Race #4 - Parking Lot Nerds, August 1st
Jun 25
DIY Robocars via Twitter
RT @SmallpixelCar: #CUDA implementation should be easy, because most of the time is on distance calculations berween two points and if the…
Jun 25