Zenmuse X5R Camera - Yielding Mediocre Results


We’ve been experiencing some questionable results when processing the data taken with our setup.  We’re currently using a DJI M600 w/ the Zenmuse X5R coupled with the 15mm Lumix G lens.  9 GCP’s were placed on the 10 acre site, at a spacing of roughly 400 ft.  Ground control is reliable and placed by our in-house surveyors so we can rule that out as a variable.  We’re using the MapPilot app to control overlap and flight lines, and feel that this app is pretty reliable.  It also controls image attributes such as shutter speed and exposure.  There’s not a terribly pragmatic way to maintain full control over the camera when flying a grid.  

During processing, step 1 was run on Full Image size.  Ground control was tagged in all available images in the Ray Cloud editor.  Camera parameters were optimized.  The project was then reoptimized and the quality report yielded all green.

In this smaller dataset of roughly 250 images, the calculated positions of the ground control are very tight (mean RMS of 0.01), but the Z value of checkpoints used can vary as much as 0.7 or 0.8ft (some as close as 20 feet from a GCP).  That sort of tolerance won’t work for the standards we’re trying to achieve.  It seems as though the processing yielded a false positive.  This is similar to the 3+ times we’ve rerun this batch of images.

My colleague and I believe the root cause is the camera, but it’s hard to say.  

If anyone uses this same camera, or has faced a similar issue with the Zenmuse line of cameras, I’d be very curious to hear your experience with the entirety of the process.  I’m open to advice on image acquisition.  Also curious to hear if I may be missing a step in processing.  



Hi Hayes,

From the screenshot we noticed that you work in feet and that the accuracy of your GCPs is set to 0.02 [ft]. This value corresponds to default values that are set for projects with units in meter. At the moment the software does not converts this values automatically and users need to manually change it to values that correspond to the actual accuracy of GCPs, e.g. 0.06 [ft]. You can do this in the GCP/MTP Manager.

The camera that you are using has a rolling shutter. Try to process the dataset using the Linear Rolling shutter instead of Global Shutter as described here: https://support.pix4d.com/hc/en-us/articles/208460436

Let us know how it goes,

Hayes, what is the latest with your situation?  I have the same setup as you but I haven’t ran it through the Pix4D process yet.  Thanks

We are using the same camera on an Inspire 1 Pro setup.  I am not sure we have seen these kinds of errors inside a well controlled area like you have described.  A couple of questions:

  1.  Are the check points inside the area bounded by the control?  We have seen these sorts of errors on the outside of a particular controlled area (i.e. checkpoints outside the perimeter of the area covered by the control points used, on the edge of a project area or model).

  2.  The other area we have had problems is on plain, uniform surfaces (concrete or asphalt), as the software has difficult modeling these well.  There errors we have seen in these areas are not big as you are noting though…

  3.  Double check the survey control.  I know you said it is good, but its always worth a look.  I am a PLS, and this is always the first place I look.  One of our projects looked good on the control, but turned out to be an issue with one of the RTK control points being blown.

Good luck, and let us know if you figure this one out.