Scanning an axial - symetrical model

Im having problems trying to scan the objects from the photos

The object Has 4 walls and every photogramrtry software i used scans only a half them. Either rest of the cameras are not detected, or are just detected as the other half, the resault of which is a poor quality scan of the half of the model.
I have only tried the rotating table technique. Can you please give an insight or point me into a right direction?

Beat regards and thank you on advance :wink:

Hi Tomasz,

-If the images you show here are the one you are using for SFM, the light is not very good. Try to have something more homogenious and if it’s reflective, do use polarise filter or very diffuse light.

-You can put your model on a flat and very patterned surface. Basicly take a newspaper, put on a flat surface and place you object on it. The camera will use this support to match as well which will help. Since it is a flat surface supporting the object, you can easily erase it in post processing.

-Use sequential matching. Not all algorithm enable it especially the commercial software that have only one matching process with parameters. Sequential matching only match the x images in a series. VisualSFM and openMVG allow that.

I think option 2 with the additional surface might be the best shot here.

Best of luck

1 Like

Thanks for a quick reply :wink:

Theese are not my photos for the process. The ones used in the software are well lit and Sharp.

Thanks for the input, ill try 2. And 3. When i gry back from work.

Also, can it be that im using a rotating table against a solid white walk ? I was thinking maybe going around the object with the camera with a changing background might help ?

hi @Tomasz.Kujawiak . Pretty cool object to scan, and also pretty difficult i think because of the reflectivity. The object looks perfectly symmetrical, maybe this is giving the half result. Can you share a screenshot of the aligned/thin point cloud?

I think that the newspaper option would be helpful to ancor your object and capture all sides. Make sure the newspaper is also enough in focus.

Your last question concerning the background: Best thing to do is mask out the background of your photos. You can do that in photoshop or, - if you use Agisoft - use a photo (without object and newspaper) and generate masks from background. Also Agisoft: You could also try to take a photo without the object (and newspaper, should you use that) and use that photo as your mask. You should then use the ’ Apply masks to tie points’ option.

1 Like

Im using zephyr so i can make a mask, but the earlier sugestiom about sequential marching also makes sense to me. Also, as you pointed - the symetry part is most likely the issue. Thank you for response :wink:

It can help not to use a rotating table and use the background as reference. However, it will be more difficult to then control the lighting, you will have to move the camera a lot etc…Plus background can also be an issue (window, plain wall…).

Additionally you can encounter problem with focus.

If you object is thinner and longer, you might need a kind of chamber rather than just a support or turning around your object.
Here since the object is a “base”, you can have the support always in view. If you don’t have the support in view for XY reason, then it can be an issue for details. For example, if you do a close up on the “leg” without a reference in sight, the software might not know which “leg” it is. But close up are not necessary anyway in this case (overkill)

It is a bit late answer… Have you found a solution? If not:

IMO it should work

  • walk around the model
  • place the object on something with a pattern
  • when you finish turn the object and redo the work. You will treat sets as separate scans
  • use a moving source of light (seems to be dodgy but worth to try)
  • delete unwanted parts from every “set” and merge results