Thanks for the reply.
I am quite new to Luxonis systems (currently working with realsense equipment), and I have a few questions. On OAK systems, is the reconstruction done at the camera level or at the HOST level?
In the first approach you propose, would the reconstruction be carried out in one of the cameras or in the host? What do you mean by "then link one OAK stream via host to the second OAK"? I am not looking for low latencies in the reconstruction, in fact the reconstruction could even be a posteriori.
What could be the approximate latency in the reconstruction in this configuration?
On the other hand, in the second configuration, calibrations would be carried out at the camera level, and the reconstruction at the host level. Can any Luxonis SDK be used to perform the stereo matching part? Or should it be done in opencv?
Thanks for everything.