Scaled Muli-Camera System

[Relyew]

Hi all,

I use a stereo camera (ZED) system to record multiple images, therefore the relative orientation between both cameras is fixed.
Given the relative orientation of both cameras (precalibrated baseline and rotation) my subject is to include this knowledge into the photoscan model.

Now my question is how can you achieve a scaled point cloud given this setup and take the fixed relative orientation into account?

So far I organized the project folder as a rigid camera system:

|-stereoCam
  |-left_images
     |-01_image.png
     |-02_image.png
  |-right_images
     |-01_image.png
     |-02_image.png

Next steps are:

• Workflow > Add Folder --> add parent directory 'stereoCam'
• In the dialogue menu select 'Multi-Camera System'
• Go to Tools > Camera Calibration --> set both cams as Fisheye since field of view is 110° (my fault, actually it is vertical fow = 47° and horizontal fow = 76°)
• Go to Tools > Camera Calibration --> load calibration file for left and right camera (parameters are given by ZED) 
• Since the offset between left and right camera is not neglectable (120 mm) we have to set a 'slave offset'
• Go to Tools > Camera Calibration > 'Slave Offset' Tab --> check 'Adjust Location' and set X reference to 0.12 [m] as well as all given relative orientation params

Current results do not look reasonable and the final point cloud is not scaled according to the baseline.
Can anybody provide me help? Thanks in advance.

In order to get a scaled model I also tried to load each camera (left / right) and added scale bars between them which actually works.
But this method does not take the fixed relative orientation into account and I really want to exploit this information.

Btw. I'm using Metashape 1.5.1.

[Alexey Pasumansky]

Hello Relyew,

Can you please share a small image set consisting of the overlapping images to support@agisoft.com and the project file that you have create following this approach (re-saved in PSZ format)?