Orthoplanes using a combination of terrestrial and aerial data

I am having a hard time getting good results from a project where we combined terrestrial, grid and point of interest data into a decent orthoplane projection.

We are following the processes outlined in the multiproject processing, and in creating the orthoplane projections. But we’re consistently getting problems where the final images are not “connected” correctly? I’m not sure how to describe the problem. It’s as if the program is choosing at random which photos to use to acquire the data.

We managed to bash together something that looks decent for one elevation, but then experienced a crash, then loss of data. I have a couple theories and questions that I’d like to run by the community because I’ve run up against the wall of my knowledge.

Here’s a link to the 3D model of the project. https://sketchfab.com/models/f98df51aea794496a2b3325f29c2e031 (you can see that the 3D turned out okay…)

This is a look at a “good” projection. Even though there are some issues this is acceptable for now. (the future will be better)

This is a look at a “bad” projection.

Detail:

The main issue in these is the transition the images are making between the Terrestrial and Aerial image data… so darn confusing, and the ghosting.

I have a couple theories about what’s happening, but really I just have these questions.

  1. Is there anyway to tell Pix4D to utilize certain images over others for the orthoplanes? Like a prefer terrestrial images… something? So I can avoid the really bad transition line?
  2. Is the image fidelity related to the quality of the point cloud?
  3. Why, if an object is _Masked _or Carved out from an image would the object still show in the Orthoplane?
  4. Does a higher number of MTPs affect the quality of an orthplane reconstruction (specifically looking at the wavy building cap in the “good” reconstruction)?
  5. Would I get a better orthographic projection by having a super high quality 3D model constructed, and then rendering out the elevations? (this means I don’t get a “to scale” elevation, but I’m looking for good quality, not to scale).

So, I’m not sure what other bits of information to add to this post, so I’ll just let some questions come to me. This is a project we’re trying to turn around fast. I’d be okay with just getting to the level of quality of the “good” facade.

Thank you

-AB

Hi Andrew, 

If I remember well, the points that are projected on the orthoplane are a weighted average of the images that see that point on the facade. The higher weight will go to the image that has a more direct line of sight to the point, as it is supposed to better see the object. That seems to be what creates the issue you are facing with the line that separates terrestrial and aerial images, where at some point aerial images see the object better, but the quality of the images seems to be less good.

  1. Is there anyway to tell Pix4D to utilize certain images over others for the orthoplanes? Like a prefer terrestrial images… something? So I can avoid the really bad transition line?

There is no such feature I know about. To avoid the transition line I would recommend the image acquisition plan I commented on your other post

  1. Is the image fidelity related to the quality of the point cloud?

The point cloud will be used to generate a Digital Surface Model (DSM) of your facade. The points from the images will then be projected onto that to create the orthoplane itself.

  1. Why, if an object is _Masked _or Carved out from an image would the object still show in the Orthoplane?

If the object is still there after a Mask is used, it could be that part of the object was not masked in the image. In this case, other images with a different point of view should also be masked to completely remove the background. Do you think this could apply to your case? See point 7. of this article.

The Carve annotation is meant to clean the densified point cloud, but does not stop the content of an image of being projected on the orthoplane. 

  1. Does a higher number of MTPs affect the quality of an orthoplane reconstruction (specifically looking at the wavy building cap in the “good” reconstruction)?

Adding Manual Tie Points (MTPs) can improve the reconstruction of your point cloud and might have an impact on the orthoplane itself. I would also check if there is noise in the edges of the building that you could remove by using the point cloud editing tool

  1. Would I get a better orthographic projection by having a super high quality 3D model constructed, and then rendering out the elevations? (this means I don’t get a “to scale” elevation, but I’m looking for good quality, not to scale).

Not sure what you mean here. 

Something you could try is to create a project where only images that see the facade (but that still have enough overlap between each other) are used and try to generate the orthoplane again. 

Hope this helps!