View my account

3DGS from 3D180 source?

Follow
Peter Graf

Hi team (congratulations,) whilst I do not have an Intel stereoscopic camera, I do have a modified Qoocam3Ultra which is now a 3D180VR modified camera shooting 2x 6944x6911px. I was wondering if your software can create 3D Gaussian Splats or camera poses towards such a process from such a rig, or if this development will be limited to your own hardware, and how the precision of such a scan would be?

Thanks for your consideration

0

Comments

2 comments

Date Votes
  • MartyX Grover
    • Edited

    Hi Peter Graf  It is possible to merge ("stitch") multiple images together into a single combined one using RealSense software.  As you do not have a RealSense camera and are also using a very high resolution though, you would likely have to use the RealSense SDK's software-device interface to define a 'virtual camera' in software that matches the configuration of your real 3D180VR camera.  It should then become possible to feed data from that camera into the RealSense SDK. 

     

    There is an example program for the software-device interface in the C++ programming language here:

    https://dev.realsenseai.com/docs/rs-software-device

     

    It would then be necessary to add further programming to the software-device application to stitch the images from the 3D180VR together, which is possible but difficult.

     

    If you had a RealSense camera, you could use the SDK's C++ rs-kinfu tool to move a single camera around the scene and progressively build up a 3D 'point cloud' image in real-time and then export it as a file once completed, instead of capturing the viewpoints of multiple cameras and stitching them together.

    https://github.com/IntelRealSense/librealsense/tree/master/wrappers/opencv/kinfu

     

    Another approach with a RealSense camera would be to use the SDK's RealSense Viewer tool to create multiple individual point cloud snapshot files of a scene from different viewpoints (which could be done with a single camera) and then use a free tool called CloudCompare to stitch these individual files together into a single combined one.

     

    There is also a pre-made RealSense-compatible commercial software option called RecFusion Pro, which can capture a scan of a scene using multiple RealSense cameras in fixed positions.

    https://www.recfusion.net/

     

    In regard to precision, when using RealSense cameras there will typically be an error factor of around 2% or less when using the camera at a distance of 4 meters from an observed object.  Error reduces as the camera is moved closer to the observed object.

    0
    Comment actions Permalink
  • Peter Graf

    Thanks Marty, this is all very useful, I appreciate your depth of reply! I shall make some further investigations ;-)

    0
    Comment actions Permalink

Please sign in to leave a comment.

Didn't find what you were looking for?

New post