Would appreciate some tips as after a week of trying photogrammetry, I have 0 results

Hey guys,

I am new to photogrammetry and using a samsung S8+ with the Snap Camera app to do some basic photogrammetry, I am trying to scan a beam of wood, its about 15cm x 15cm x 60cm and im having pretty big issues with it.

I need all 6 sides to make the asset usable fyi.

I have scanned it in 5 environments already to try and get even and enough light as some of the earlier attempts didnt have optimal lighting, I have been making sure there is no blur in the images and the ISO and shutter speed is good. Unfortunately, no matter what I do, I get the same issue. Each of the long 4 sides roughly matches up, as I am taking photos of each corner side too, but the end ‘caps’ just never work, they either line up properly and dont texture, or dont line up at all, Ive tried to get sort of an ‘arcing’ coverage from the side down below to try and help my apps realise that the bottom is indeed attached to the sides, but it just never works.

I have the beam hanging from a nail to ensure I can get all 6 faces, and the beam is not moving when I am doing the photogrammetry.

I have also tried control points but again, same issue, it just wont have it.

Ive tried reality capture, meshroom and agisofts photogrammetry software, and I get varying results but all are sub optimal.

Does anyone have any suggestions on how I can do this, because I cant progress any more.

Thank you in advance for any help.

Hmm, maybe it has to do with the camera calibration of your phone, in the photogrammetry software. Or maybe the photo taking strategy is wrong. Can you make your best photoset available, maybe i can see some obvious errors.

1 Like

Hi @Hambo - welcome to the forum!

As @cie says, sharing some of your image data (ideally the whole set) via flickr or something would be useful in helping diagnose the issues.

You can normally get a fairly usable set of images by just laying something like this beam flat on a table or the floor under even/diffuse lighting and capturing one side, flipping it over and then capturing the other side.

This workflow works best if you are masking out the backgrounds of your input photos in something like Photoshop before loading into photogrammetry software or using built in masking tools in Metashape or 3DF Zephyr.

@cie @nebulousflynn

Thank you so much for offering to help.

I found a fix after reading your messages. I tried out moving from my Samsung S8+ to my Canon G7X point and shoot and insantly got results, it seems that the increase in quality going to a true camera has just allowed Reality capture significantly more opportunity to align images and get the job done.

Again, thank you so much for taking the time to actually reply and offer to help, I appreciate it massively…

And thank you for the welcome Nebulous!

Have a great week!

2 Likes

Glad you figured it out :slight_smile:

I use a Canon G7X a lot myself!

A few tips:

  • shot RAW if it is possible
  • take more photos that you need
  • if you use shallow DOF don’t build texture from photos, use vertex color instead
  • check performance of your lens when aperture is smallest possible. Some lenses have quality that low that it is better to open aperture and take one or two steps back
  • use equally exposed photos, it means you should optimize brightness at the level of developing RAW or you should change exposure settings during shooting (it is tricky). If no, you can finish with no data in shadows and when changes in exposure are small, software normally do the job by averaging the photos
  • shoot in natural light
  • use tripod. The photos shouldn’t be blurry. When you are trying to take a photo from an extream angle it is easy to lose sharpness.
  • tweak with parameters in your software. Even really bad set of photos can be saved. Sometimes it isnt worth but if you cannot redo shooting, don’t panic.
1 Like