500 image limit for structure processing?


#1

I frequently take large scale images often over 500 images. What is the reason for the 500 image limit when trying to process something as a structure and can’t be flown at less than 30m?

Is there the opportunity to look at increasing this in the future?


#2

I have experienced that you can actually take too many images when trying to model structures. On the occasions that I have noticed this the program has had problems stitching together the faces of the walls and sometimes you even get a ribboning effect. Unlike terrains you only need enough points to construct larger flat faces. The lesser the amount of images that the program has to use to stitch a wall the better that wall is going to look. In my opinion you don’t need pictures any closer together than about every 3-5 degrees. That’s 72-120 shots per orbit. The current 3D mode “orbit” takes way too many pictures.


#3

I did this one with 85 oblique images flown manually at two altitudes. No verticle images.
Sculpting 3D models with a drone is part art and part science and the technique must be adjusted to the subject matter.


#4

@Gary, love it! Thanks for sharing.


#5

Hi @Andrew1,

That’s a great question! We limit all plans to 500 images processed with Structure mode because our processing algorithm is not designed or optimized for large datasets. Processing under Structure mode is very computationally intensive.

We’ve run tests and concluded there is a diminishing return for datasets larger than 500 images. More images tend to be problematic for our processing engine and it will take long longer to deliver the results. We cap the limit to 500 to ensure a higher processing success rate for our users and we currently do not intend to increase that limit. I hope this explanation helps!

Best,
Stephanie


#6

Thanks all for the replies. It seems my theory of more detailed photos the better isnt quite right.

Ive attached the map in question which consists of 351 images. I flew a mission at 39m and screwed it up and had errors in the processing with these towers of blur and pixal all through it.

I flew is again at 20m thinking that it would capture more detail but i couldnt upload it in Structure because of the altitude.

f0a9230340_ANDREWNEIOPENPIPELINE by DroneDeploy on Sketchfab

This is the result in uploading it in Structure from the original 39m flight.

eff6bf3f7f_ANDREWNEIINSPIRE by DroneDeploy on Sketchfab

As you can see it didnt turn out the best hence my reason for trying it lower…which is clearly wrong

@Gary I and starting to understand your comments about 3d modelling.


#7

You can see the ribboning in the first one. The second one was actually closer to a great result as you can see the open space under the railway. Did you run any nadir images on this one? There’s allot of horizontal detail in this so I would recommend it. If you can fly it again try 55/60 overlaps (you don’t need too much detail here) and setup two orbit missions in DJI a third and two-thirds down the rail.


#8

Thats everyone, ill give it another go and see how I go.

Thanks again


#9

For a structure such as this, could you fly a nadir initially and the oblique after? ie Can you combine both styles of imagery when processing in 'structure 'mode or is it best to keep all images at some kind of oblique angle?


#10

@TheAeroScout Yes, nadir and obliques in the same processing session. Since this post they have released the new Structures Mode which does what I did here. The part to figure out with the new mode is how far your perimeter needs to be offset from the face of the structure. They fix the gimbal at about 60d so the math can be done in relation to your altitude. Try 60’ out for every 100’ high.