Agisoft Metashape

Agisoft Metashape => General => Topic started by: FLuca on December 04, 2017, 01:46:11 PM

Title: Issue when scaling model with camera relative postions
Post by: FLuca on December 04, 2017, 01:46:11 PM
Hello Guys,

I am confused with poor result in scaling 3D models.

Instead of taking reference measurements on the model, which is not very precise on human body, or marker on the ground, I wanted to use the camera relative distance.
On my rig this distance is fixed and larger than any feature that can be reconstructed so I expected to have more precise scaling than using object feature.

So I scanned a wood chunk 8x8cm large with a camera distance of around 2 meters from each other.
The result is very disapointing. The resulting model is about 8.5% larger than reality.

Even with unprecised measurement of the camera positions, I expected only few %

My questions are:

What does the camera position represent in photoscan: focal point, sensor center, ... ?

Did any of you used this method with success ?

Any tip welcome


Title: Re: Issue when scaling model with camera relative postions
Post by: n@sk on January 08, 2018, 03:02:14 AM
this is a month old now but I hate to see unanswered questions...

if you used the camera base in the end of the process, chances are that the result will be disappointing.

Have you tried
import cameras->create scale bars->define reference distance->set scale bar accuracy to ~0.001m->align->optimise

it represents the perspective centre
somewhere around the middle of the compound lens system which means that you would have to calibrate your rig (using markers on and above the ground and an actual scale bar) before fixing this distance
Title: Re: Issue when scaling model with camera relative postions
Post by: FLuca on January 08, 2018, 12:23:08 PM
thanks for your answer

after several testing I found that the camera position in photoscan is corresponding to the entrance pupil of the camera/lens

when using the figures as defined in the nodal ninja database https://wiki.panotools.org/Entrance_Pupil_Database (https://wiki.panotools.org/Entrance_Pupil_Database) I found that the measure distance is correct

but as you mentioned, I now take a reference shoot with a visible scale and use the value for the other shoots