With Ground Control Points, Less is More
One of the best ways to increase the accuracy of a drone map is by including Ground Control Points or GCPs. A GCP is a physical object or painted target in the scene with a known position. GCPs are typically located by a surveyor and easily seen in images taken by the drone. Bright contrasting colors with an easily identifiable center make the best GCPs. Operators can identify the GCP in images using structure-from-motion (SfM) software such as ContextCapture, and located them on the earth with surveyed coordinates. It's helpful to think of the drone map as a rubber sheet and the GCPs as pins that position it on the earth. There is a temptation to put as many GCPs as possible in the field with the hope of increasing accuracy. However, in my experience, less is more when it comes to the number of GCP. Every time you add a GCP, you're providing the software locational data, but you're also introducing human error. With each image you tag with a GCP, you may be locating the center 1 or 2 pixels off. You're the human, so the software is going to assume you've positioned it correctly.
I'll give you a real-world example. I was mapping a 4-acre site. My map was a meter off when I included zero, one, and two GCPs. Once I added the 3rd GCP, I was within 2 inches, and when I added a 4th it was sub inch. However, after the 4th GCP, the error continually increased by approximately a tenth of an inch for every GCP I added. The reason was that every time I added a GCP, I was also adding human error. I wasn't locating the center of the GCP perfectly, and the software was "correcting" around my error.
Different software provides different recommendations for GCP spacing. ContextCapture recommends GCPs should be spaced no more than 20,000 pixels apart. That would be every 500 feet for a Phantom 4 Pro flown at 100feet AGL.