Issue while trying to do a body scan

[Narvalo]

Hello there

I am new to Photoscan and I am currently trying to do my first body scan. For that, I took almost 80 photos of my model, outdoors. But as my terrace was too small to turn around him, I asked him to simply turn on himself, the camera remaining still.

For my first try, I tried directly to scan by aligning the photos, without masks. But I got a cloud of very bad points. So, for my second try, I cut out my model on each photo with Photoshop, then I imported these new photos in Photoscan, this time applying a mask to remove the green background. But again, here's the cloud of points I get: (see image 1)

The parameters I used to generate the cloud was: (see image 2)

I really don't understand why I'm getting such a bad quality of points, the cloud doesn't look like my model at all.

If you want to try by yourself with my photos, you can download the photos here :

https://mega.nz/#!ecISVAhT!P674GEkim7YvYKtOvFgsKolJxWEoQuCUDPKd7HPmoVk

Anyone could help me please ? (And be indulgent, I am a complete beginner !)

Thanks by advance !

[chrisd]

It is likely that small movements of the subject between each picture are throwing off your results. People move even when they try to be still. The arms extended pose makes it harder for the person to remain completely still.

I have seen some reasonably good results from a person that was seated using only a single camera and moving around the subject.

[SAV]

Hi Narvalo,

High quality body-scans are only achievable using a multi-camera setup/rig where all cameras are triggered simultaneously. Look at this: http://www.pi3dscan.com/

If you told your model to turn, the small changes that occur during the rotation will not allow for a proper reconstruction because the object is slightly changing from one picture to the other (e.g., leg position, arm position/angle, etc). The SfM algorithms that are used in PhotoScan are not able to identify key points if the object is changing, resulting in an untidy/useless tie point cloud.

Regards,
SAV