bokeh-camera

If you think the Google/Zoom segmentation-based background blur effects look kind of fake, this attempts to simulate a more realistic-looking bokeh using a RealSense D455 (for best results) or D435 camera on a virtual camera device that you can use from any conferencing software.

Work in progress, code not yet cleaned up, but roughly:

1. Build requirements:

   • librealsense2-dev
   • libopencv-dev
   • cmake
   • g++ (or similar C++ compiler)

2. Run ./build-this

3. Ensure that you have the v4l2loopback kernel module

   • Ubuntu: sudo apt install v4l2loopback-dkms
   • Fedora/RHEL: sudo dnf install v4l2loopback

4. Set up a v4l2 loopback device at /dev/video20

   sudo modprobe -r v4l2loopback  # This one can fail if the module isn't loaded, that's okay
   sudo modprobe v4l2loopback devices=1 video_nr=20 exclusive_caps=1
   

5. Run `./BokehCamera

6. Select the virtual camera on Google Meet or Zoom or whatever you use

Controls

While it is running:

Press 1 to send bokeh-fied image

Press 2 to send a map of the computed amount of blur (white = don't blur, black = full blur)

Press 3 to send the depth map

Press 4 to send no-bokeh RGB

Press [ or ] to refocus the image

Will it work on a D415/L515/T265?

Probably not but go ahead and try, hack at it, submit a pull request.