The oblique imagery is for higher quality in the z axis and to render steeper elements such as cliff faces and buildings. If only looking for a good ortho then nadir is fine.
More GCP should be placed around sudden elevation changes, spread them over the entire area with a maximum spacing of around 100 to 150m.
Fly beyond the area of interest by at least 1 image width
65/65 produces about 12 images per point. I ran 75/75 yesterday at 55deg and got an average of 30 matches. A shallower pitch on the gimbal will get more FOV and the data at the edges of the frame can get distorted, but you get more matches and mass shapes get captured better. I was just capturing a 3D model of a construction site with allot of heavy equipment and trees.
Main take-away, if needed to be precise within my usual margin of 1-2 cm (which I’m spoiled with, from doing 0.7 px/cm GSD flights doing volume measurements), I need to place a shitload of GCPs to untwist the model completely…
Conducted a few test with obliques at -60 deg (at 70 meters) and nadirs (at 100 meters), resulting in 2.6 cm/px GSD. Placed GCP in a straight line more or less, with a single outlier perpendicular to the main direction, 50 meters away.
Visible warping, but that disappeared after starting to clean-up the thin-cloud. (decimated from 800k points to 300k ish).
Another learning was that mature barley with its soft strings is a bi… to find tie-points on when the wind blows with 8 m/s. I flew a few hours earlier in 5 m/s, many more tie-points, and few days before that in 3 m/s I had no problems with tie-points Rapeseed, on the contrary, had no issue what-so-ever.
Precision wise, using 14 gcp’s spread over 1 km wasn’t impressive.
For the GCP’s, avg 3D RMS of 3.1 cm, worst GCP being 4.8 cm in RMS error. I had higher errors previously (up to 12 cm), but re-ran the alignment after having all GCP enabled, which took it down to the above.
I even included a scale-bar for checks. over ~76 meters, it had an error of 12 cm. Measured with a 100 meter steel tape-measure, it is usually within 2 cm at those distances (have used it quite a lot on other projects).
Measuring a 20 foot container in the pictures also has a similar error, measuring 5.96-5.98 m when it should be 6.06 m.
But I guess that is what one gets with higher GSD and a very long narrow model.
Yes, crops are a bi… What time did you fly? I agree 4.8cm RMS is way too much so that tells me that the stitching and point detection struggled badly. How large was the total area?
Did you fly a higher overall as well? This usually helps with homogeneous subjects.
Not saying that PPK would have been much better finding tie-points, but the horizontal dimensions should be much tighter. RMS should then drop to the 1cm level you are looking for.
If I have to fly something like this I always try to do it with the sun straight up so shadows are eliminated as be as possible. This will also help get ground points if there are any gaps.
What angle were your obliques at? I may not have mentioned here, but obliques can cause problems with GCP’s.
I don’t think you can go wrong trying to PPK… Thanks for the detail! I am sure it will help many others.
Tried post-processing the logs, and might have found the reason, and actually quite impressed with RTK-engine of the RS+ for keeping the fix for most of point collection !
I stuffed the rover-pole and mounted RS+ in my car, so 1 meter of the pole was sticking out of the rear passenger side window, with the pole angled 45-50 degrees. Then proceeded to drive on a gravel road. So my guess is that the angle+vibration got the better of it. The “static” points themselves seem ok (1 minute collection time) but the surrounding points (during driving) are both float and single.
Gotta push the gear! Now for working out a method of mounting the rover-pole on the car that doesn’t cost a fortune like some Seco solution I’ve seen
Rapeseed, on the contrary, had no issue what-so-ever.