Video encoding scheme for smart doorbell [on hold]

_x000D_ _x000D_ I am developing a smart doorbell using RPi Zero W and Jetson Nano. The Zero is the core of the IoT appliance ( the doorbell ), while the Nano is the hub and runs the face recognition algorithms. So, what I am unable to choose is the way to send the camera video stream from the Zero to the Nano. I am thinking about Websockets or WebRTC, maybe streaming the video with MJPEG. But what about VP8, H.264, etc...? I understand that, if I choose one of them, there will be more work to do ( to encode/decode ), so it is a burden performance-wise. But what about feeding the neural networks? MJPEG gives me a bunch of sequentially ordered images, so I can send, step by step, the images to the net, one at a time. So are there other nuisances introduced by video encoding schemes in my use case? Because I need to stream ( on demand ) the camera to clients ( smartphone apps ) and because the clients can be everywhere on the internet ( non necessarily on the same local network as the doorbell and hub ), I think MJPEG could be inadequate. What do you think?