15% Difference between Pix4d volumes and manual survey


I am using a trial version of Pix4D and a new Mavic 2 Pro drone to survey stockpiles. So far, as described in the subject line, the results are unacceptable.

I have flown 2 surveys and in both cases the difference between the drone volumes and manually measured volumes is over 10%. The manual method is considered accurate, so I need to either improve my drone surveying/processing technique or convince people that the manual method wasn't that good after all!

The surveys were actually flown with DroneDeploy, as Pix4DCapture doesn't work for me (another post some other day). I used the Overlap Optimizer tool to ensure adequate coverage. The parameters were 80 m high, 75% front overlap & 65% side overlap for the first survey resulting in 232 images,  and  60 m, 76 % & 73 % for the second, with 178 images. The stockpiles are 10 to 15 m high.

I also surveyed some of the piles manually using a Topcon/Magnet system. Sadly these are done with local coordinates so it is impossible to merge the two systems for a better comparison.

The Pix4d  processing is completely default. I am using the triangulated option when defining the base of the piles. I picked the boundaries from the volumes view. I found it useful to rotate the perspective as I went to visualize the edge of the pile.

For the first survey I obtained volumes of 4850, 1907 & 3574 m3 manually and 5544, 2260 & 3856 m3 with the drone, which corresponds to differences of 14%, 18% and 8%. The results of the second survey were similar, but the manual survey was done by someone else.

I can understand and expect the drone survey to be somewhat higher because of increased resolution but not this much.

My questions are...

  1. I assume other people have done similar comparisons... what is a typical difference?
  2. Its hard to ensure I'm picking the boundary correctly. It is sometimes quite difficult to spot  the edge of the pile. Any tips, pointers, docs or videos I should watch?
  3. Are the default processing parameters suitable?
  4. I would find it a lot easier to pick the boundary points in the orthophoto which is much more detailed. Any way of doing that?
  5. I've exported a boundary to a shape file, imported to qgis and then reexported and the volume went from 2260 to 2166, which is kind of amazing. A detailed description of how to do that process correctly would be very useful.




Please sign in to leave a comment.