Question with creating an ortophoto

[User7583]

Dear Friends!
I have tried to create an ortophoto texture many times but always had bugs on it.
Could you please look throuh my files and advise what should I do to gain a perfect texture!

I'm ATTACHING the pieces of texture which show bugs. The photos are'n the best ones but I'm considered that that's not the reason for the bugs. Photos are made with Sony a5000 with 16mm lense on the height of 100 meters.
The texture: https://yadi.sk/i/39qjK3zUhY4UN
The bugs:
https://yadi.sk/i/-PTUqmsBhY4Wc
https://yadi.sk/i/z9kDT8dOhY4We
https://yadi.sk/i/-vx0cQtchY4Wi

These are the explict steps I did, creating the texture:
1. I added all the 42 photos

2. Align photos
   accuracy: high
   pair preselection: disabled
    Advanced:
   key point limit: 40000
    Tie point limit: 10000

3. Build dense cloud
 Quality: high
 deph filtering: aggressive

4. Build mesh
  SUrface type: heigh field
  Source data: dense cloud
  Face count: high (at about 6,6M polygons)
 Interpolation: Enabled (default)

5. Build texture
  Mapping mode: ortophoto
  Blending mode:  mosaic
  Texture size/count: 14000 x 1
 Enable color correction - none.


As you can see I didn't use any other tools.
I didn't use "Optimize cameras", "Tie points", "close holes", "decimate mesh"
I tried all that additional settings and variations of them and alwyas had similar NOTperfect results.

I belive there is a simple thing that I just dont get!
Please help!
I'll be very gratefull if anybody helps!

[stihl]

Hi User7583 (charming username)

As your goal is to generate a pleasantly looking Orthomosaic there are a few things you should know.
First off; what you are seeing aren't called bugs but graphical artifacts. The cause of them are irregular and or incomplete height data in those particular areas (at the cars and trees). Those areas are incomplete  because Photoscan was not able to determine the relative exact height data needed because of occlusion and shaded areas or because of low overlap.

The (True) Orthophoto that Photoscan produces is generated from the dense point cloud and mesh. If there is noise or incomplete data in the dense cloud, then these graphical artifacts will be shown in the (True) Orthophoto.

Thus it is important to generate a clean mesh. Preferably a mesh that's generated solely from the ground points and not all the points.
You can use the automated ground classification that Photoscan has to create a classification of only the ground points which you can mesh into a model.

Artifacts in Ortho's is a problem with Photoscan. Version 1.2 supposedly will offer seam line editing which you can use to create an aesthetically pleasing Ortho which retains it's georeference.

In short; you can try the ground classification to classify only the ground points, then generate a Mesh from solely the groundpoints (In the Build Mesh window you can tell Photoscan to only mesh certain classifications like ground points) which should in theory minimize the artifacts.
Although I'm not sure about the actual output which this should give as I haven't tried this in a very long time and I have forgotten (I've moved on to different software for Orthos)


Your workflow seems to be good. However you can try using 'Generic' for the pair preselection for the alignment as it should decrease time needed to align.

Before step 4, build mesh, you should use the automated ground classification.

I can also advice against using a *High* dense point cloud, especially if your goal is to make an Ortho.
As High will generate a 3D point for every 2 pixels in every paired images, you will end up with an oversampled Dense cloud that's difficult to work with (because of the sheer size and density) ..which will in turn also increase the time needed to classify the ground points.
Also for generating an Ortho you do not need to creata a texture. These are meant for 3D models and not Orthophotos.

[User7583]

Thank you Stihl for your answer very much!
Now Im trying to classify points of dense point cloud just how you said but "Classify ground points" is unactive. Why is it so?

[James]

I believe classify ground points requires that your chunk is referenced either by camera or gcp coordinates, so that it knows the scale of the scene and which way is up. You will see an [R] symbol next to the chunk name in the workspace pane when it is referenced, currently in your screen shot it is not.

[User7583]

Thank you guys a lot! I'm working on it an will share my results.