Simulated 3rd Camera with stereovision
Hi all. I'm currently researching the possibility of simulating the stream of a simulated camera from two physical cameras in close proximity to the simulated camera. For example, 2 cameras on either side a gun barrel, feeding a stream into a processor to generate a view from inside of the gun barrel if the barrel does not exist. (Iron sight/scope compensation for the bullet drop over distance is irrelevant for now) I haven't done any coding since freshmen and doesn't have a clear grasp of what it takes to make such device/algorithm work.
I know 3D reconstruction of polygons and rendering them again from a different angle is resource intensive. I would prefer a solution that have latency within several milliseconds. Neural network sounds ideal in theory, but with millions of pixel data points as input and millions as output, implementation and training of such a network is beyond my comprehension.
I know OpenCV should be capable of boundary recognition producing a depth map using two camera. In theory these data should help optimize the neural network above. Anyone have any opinion or advise?