The problem with typical approaches to scaling up data collection just take a standard flight path and do it a lot of times. This compounds error over the space it’s done in, which with really large data sets results in a notable bowl in the data due to incorrect lens calibration estimates.
There are several fixes:
- If possible, fly with a fixed lens like Sensefly’s SODA (which is a repurposed DJI, FWIW)
- If possible, fly with RTK GPS
- If possible, use a global shutter rather than a rolling shutter camera
- Flight plan with calibration in mind as follows:
With flight planning, there are two options:
One is to do periodic calibration flights, such as reviewed here: https://www.opendronemap.org/2019/12/self-calibration-of-cameras-from-drone-flights-part-3/
With a flight planners like sensefly’s or QGroundControl, you can append these calibration flights as a smaller flight area within your larger flight, and then there is a calibration flight associated with each larger flight.
A simpler approach is to just vary how flight planning is done. Typical large scale flight planning is usually just more of the same as small scale flight planning like this:
Instead, rotate every other flight 20 degrees from the previous, and every other flight vary in height as follows:
So instead of having every flight with a yaw of 90 degrees, every other flight should be 80 or 100 degrees. If flying at 800 meters, have half the flights at 780 meters and half at 800 meters. Also, pitch the camera at 80 degrees instead of 90 — straight down camera views will result in self calibration problems.
Note: For simplicity of drawing the diagrams, I simply rotated each rectangular area giving triangular overlap between the flights, but the overlap can be rectangular.