Open MEBestawy opened 1 year ago
I would imagine that most places that have CCTV cameras have some desktop/ local network that they all connect to. So I think we can try to slim down our current design and have it run on the desktop rather than directly on the cameras.
We could also try an approach of prioritizing certain events over others. For example, normal activities don't need to be logged in real-time. We can take our time processing those. Whereas incidents will need to be detected immediately, thus needing a smaller model.
A third suggestion would be to use the edge hardware to just compress the videos. We then send the videos to our own servers to be processed how we like. That way is more costly for us, but it might the fastest if network delays are shorter than our processing delays on small devices.
In the pitch feedback, our partner mentioned that 3D res-nets would be too large for most cameras. What are other potential solutions other than the suggested hardware installation?