Photogrammetry with a robotic arm.

[Rbailey]

Hi all,

I'm wondering about the feasibility of attaching a camera to a robotic arm and then use the arm to get a series of images with which I can then create a 3d model of the work piece the robot will work on.

My questions are:

• Has this been done before?
• would knowing the exact position of the arm at each image increase the accuracy of the model?

Thanks for your time,

Rongomai

[chrisd]

If this will be used to predict interference/collision with the robot, its possible the scaling uncertainty would be below the accuracy you require.

Typically the scanning would be done with precalibrated equipment (ie laser or structured light system) so model accuracy will be known ahead of time.

You would need to capture clear references that are known to better accuracy than the model needs to be to minimize scaling error.

[dmccreadie]

Have a look at CultLab3D's system. https://www.cultlab3d.de/ They are using a robotic arm.

[Kiesel]

Hi Rbailey,

in principle it can be done by using a normal processed project with a robotic arm, from that export cameras in Agisoft xml-format. For every new project you have first load your new photos (must be from same positions and same directions as before of coarse) and then hit import cameras. Imediatelly after import of cameras you are prompted to build point cloud.
You can improve accuracy by involving coded markers.

(Btw. that is also possible with camera rigs as long as they are stable.)

Hope it helps

Kiesel

[jrp]

Hi,

www.photorobot.com (and other companies selling the same hardware) has a few impressive options too, even if the options are buried in thie product pages. They are primarily about product photogrpahy, but a robotic turntable with 2 axis robot arm goes a long way to getting good photogrammetry results, I know it's been used to good effect in at least one institution.