How to increase Orthomosaic accuracy

[BrianBulla]

Hi,

I'm using a DJI Phantom Pro 4 to capture imagery which I then process in Drone2Map.  I am trying to create an accurate orthomosaic using the images.  While it looks good, it is off in terms of the accuracy of it.  The image below shows my captured orthomaosaic against a 15cm accurate ortho we get from a vendor.  As you can see, the road centrelines are off by about 2m.

Is there a way to get better accuracy?  Either through an enhancement to my drone, or through adjusting the settings in Drone2Map??

[CodyBenkelman · Accepted solution]

Brian

sorry I didn't see your message sooner.

Are you using ground control points?  Without ground control, your mosaic may be off by many meters.

You get the best accuracy if you can survey points that are visible in the imagery using a high quality GPS receiver, then import the control points into D2M, "link" the control points to the images, and re-run your orthomosaic.

If you don't have independent GPS control points you can extract control from the Esri imagery basemap.  This will ensure your mosaic will align with the basemap but be aware that our basemap comes from many sources and Esri can't provide a stated accuracy for the basemap.

A third option (if inside the USA) is to use the NAIP image service to extract control.

Cody B 

[BrianBulla]

Thanks Cody,

What I ended up doing was using our existing ortho, and just using the georeferencing tools to pull everthing into place.

At the time, using control points was not an option, but going forward I will look into adding that to our process.

Thanks!

[CodyBenkelman · Accepted solution]

I should have offered that as an additional option - if you have an existing ortho base, you can load that into Drone2Map and then use it to extract control points.  

This works for horizontal control.  A simillar process works for vertical control but you'd load in your own DEM...

Cody B

[DerekPeck]

We're setting out targets on a flat floodplain and taking GPS with a sub-centimeter RTK unit and getting anywhere between 3"-6" with our control points in d2m.  Does this seem about right for horizontal?  

[beel]

Just keep in mind that adjusting to the orthobase only corrects your imagery to the orthobase imagery. You'll have the same error, so the two will match well, but it's not quite as good as georeferencing against surveyed points.

You can do the same thing with using LiDAR data to create psuedo GCPs in order to nail your 3D surface to the ground, btw.