Mapping with crap GPS

Mapping newbie here, for a while now I’ve been fascinated by OpenDroneMap and its capabilities, but I’ve suddenly realized I’ve missed an important part of mapping.

This whole time I’ve been using a DJI Mavic 2 Pro drone, thinking all was well. But then I learned about RTK/PPK GPS and realized the consumer-grade GPS in my Mavic wasn’t up to the task.

So that explains why my orthophotos come out riddled with artifacts and stitching errors, and possibly why the trees look like green, space-warping blobs.

Eventually I’ll have to spend my hard earned money on some GCPs and an RTK GPS device, but in the meantime are there any settings in OpenDroneMap that allow it to work better with cheap/inaccurate GPS?

I’ve noticed MICMACNode handles these jobs better, producing larger but less frequent errors that I can usually clean up in Photoshop (it’s fine, I’m not interested in doing survey-accurate work just yet). Nonetheless, are there settings for MICMACNode that might also improve things?

I’ve tried skipping the geotagging altogether by stitching the photos together using Photoshop, but with so many large photos and so much detail I need a bank of supercomputers to achieve this. Is there a QGIS plugin that might work better?

1 Like

For naive stitching I have found Microsoft Image Composite Editor to be incredibly competent.

I’d be surprised if your GPS was causing your reconstruction issues, unless you were routinely getting inaccuracies of like, 100m+

RTK/PPK and such are amazing, but I wouldn’t want you putting money on something that may not get you what you think it will. Their value in this is more about ensuring the reconstruction is quite literally well grounded and where it ought to be in the world.

What’s your flight planning looking like?

1 Like

Sweet! I’ll see if I can find a copy of Microsoft ICE but it seems to have been deprecated.

Re: my flight plans I use Pix4DCapture. I haven’t done a mapping job in a while so I’m a bit fuzzy on the details but from memory, for both orthophotos and photogrammetry I usually set the speed to fast, the overlap to 80%, the camera angle to 90 degrees and the height to 60, 80 or 120m depending on the job. I don’t usually pay attention to the GSD but I think i recall it being most ideal at 60 or 80m?

Here are some examples of the errors I’m getting. I only ever use ODM’s presets and don’t usually change the settings. It’s worth nothing that these errors are always far worse when I use fast-orthophoto.

ODM turns buildings into jelly, despite using the “Buildings” preset:

trees in ODM warp space near their edges:

MICMACNode turns this pond into a jigsaw puzzle:

MICMACNode causes massive rifts in the timespace continuum:

I think in most cases our pipeline should have surpassed the quality of MicMac.

As for the presets, I’m not sure that they don’t need updating and tweaking after all the many updates Piero and others have made in the past year. We’re able to do so much more than we were before.

I think, if you have not yet, poking around in the Arguments section will be super helpful.

If you can share a dataset, I can also take a crack at processing it to see what I can get to give you an idea as well.

Thanks for the tips!

When you say the “Arguments” section, do you mean “Options and Flags”? I’m afraid I’m not finding much useful there as most of the topics simply describe what each option does, without much in the way of “what works better in what situation”. It has inspired me to try using -ignore-gsd, -pc-geometric, -pc-quality and -use-fixed-camera-params, and in some cases I have seen a small increase in output quality.

Are there any options/flags/arguments you would recommend in general? I’m useless at command line stuff, so I’m using WebODM and I suspect there might be some options that don’t appear in the Web UI.

I have a theory that since I’m using Pix4Dcapture and setting the flight speed to fast, the issues I’m having might be a result of the Mavic 2’s rolling shutter distorting the input photographs, as well as interfering somewhat with the GPS readings. I’ll test that theory when I get the chance.

I’m afraid for a couple of reasons I am not able to provide past datasets.

I managed to find a copy of Microsoft ICE! I gave it a quick spin and got mixed results, but I’m sure I can get it working with a bit of tweaking.

1 Like

Yes, that section.

I’m working on adding more context to each option to address what you have found lacking.

Rolling Shutter distortion can be a big problem for fast collects over large areas, for sure. I don’t think your GPS is as much as a concern as you, however.

It’s hard to say without knowing exactly what you need/expect and what you’re working with.