I have been working on a drone-acquired imageset of a wind turbine that I’ve been trying to construct. It is devilishly hard for two reasons I’ve identified, which I believe compound one another:
- The positions and attitudes written to the image metadata are inaccurate - the residual lines are very long.
- Geometry and surface texture of the wind turbine. The surface is generally white and featureless, the blades are thin (difficult for RC to differentiate one side from the other, given GNSS/IMU problems), and inevitably, a lot of background features (terrain, sky, terrestrial objects) are captured in the images due to the generally long and thin nature of the object.
The model that I’ve arrived at is full of holes and irregularities despite copious placement of control points.
I will likely have to recapture it - though until then I will continue to try to align it using control points.
Here is my question; should I have to recapture it, should I use a strategy of:
A shorter focal length / wider field of view, using a full-frame 42MP or medium-format 100MP camera? I’m thinking flying 4 orbits, with the entire asset in the image, with 90 photos per orbit (every 4 degrees). ~400 large images
A longer focal length / narrower field of view, using a drone-integrated camera (24MP 35mm or 50mm prime), with 10-12 orbits, 90 photos per orbit, with the goal of filling the frames with the asset as much as possible.
Thanks kindly for your input, in advance.