Hi, I did not find this issue using the search function, if the topic exist please refer, thanks
P4P, alt 28m, pix4d capture, camera 75 angle, 75% overlap, speed normal. near 12pm local time. mostly sunny. flew a diagional pattern
warehouse construction, steel framing,
processed with webodm presets: buildings, forest, 3d.
I dont understand why the gsd is 1.6cm, I expect .75.
The steel is displayed on the ground, there are no walls. the steel components, beams and joist dont align correctly.
I re-flew the building with a double grid and got the same results.
any suggecstions? My opinion is that the camera is not working correctly.
Perhaps due to being processed at high feature quality, rather than ultra?
Your model might benefit from images taken from just beyond the perimeter looking inwards, at a lower height, and at -45 or 50°.
I think in difficult reconstructions like this, imaging from 2 or 3 heights AGL is helpful, making sure there are enough common features between each set to tie it all together.
I don’t think there is anything wrong with the camera.
Agreed on all points. This might be one of the single most difficult datasets to collect well enough to get good stitching because you would need to get good feature matching on the steel itself but also because our meshing approach in ODM doesn’t handle lots of holes particularly well.
1 Like
Thank you for looking, yes I will try more photos as described.
Im still nonprofit so currently duing my self training I learned to limit photo count to around 600 max to reduce processing time to 2hrs.
After studing your responses I realise the program has to deal with shadows of the steel which are at ground level as well. do you think the shadows are the culprit? at least as far as the steel?
Perhaps if i somehow marked the steel or shadows? or edit the photos in someway before uploading.?
1 Like
If you want to limit the number of photos but do a robust test, limit your area to a subset. Combining a grid pattern with orbits may help for trickier objects:
If you are doing a small enough area, then you can take a lot more photos than you think you need, and then try different combinations of them for processing to figure out the minimum you need. For example if you get a dataset that processes well with all the photos you have taken then you can process again with every second photo, with every third photo, etc. until the model no-longer works. Doing this with a very small area, then allows you to quickly iterate and find your minimum, optimized flight plan. Then you can try it with a larger area with confidence.
And: physically (not virtually) marking steel could be helpful to give it texture, but the least invasive trick here is closer photos and lots of angles.
1 Like
Great, with your advice I have a plan. look forward to flying that next week.
just put the square building in the round hole…
sincere thanks.
3 Likes
Ok, I reflew the jobsite using pix4d capture, simple orbit, I flew two orbits 1m difference altitude, plus one grid with 50% overlap, (low battery is why) and 65deg camera angle. the orbits were set to 10deg. at 24m
processed with no resize 3d model. unfortunately the steel framing was covered by deck. and your software did a nice job of not including including the photos that werent needed for 3d model.
question, I understand the documentation to state, the resize does not effect the actual output, the resolution is set by the feature quality, is that correct? what I noticed is webodm showed the gsd as .67 with no resize of photos. I dont know if that is related, or I changed some other setting.
If you can, set both your feature-quality and your pc-quality to ultra. Turning off resize is an important first step, but no additional quality will be seen without feature and pc-qualities increased. For noise abatement on a structure like this, feature-quality is critical. pc-quality will also improve detail, but the first is more important for this use case I expect.
1 Like
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.