Incentivised Multi-Target, Multi-Camera Tracking with Untrusted Cameras
medium.com
medium.com
After all, they don't have any control over the many public cameras controlled by others, but they do have control over whether or not they install a camera of their own.
I fear you've spent quite a bit of time on the technical aspects and theoretical human aspects, but will run head-on into the practical human aspects and bounce off.
I see your point but remember this is all public information gathered from public areas where there is no expectation of privacy. There are few circumstances in which you can stop anyone from taking a picture (or a series of pictures) in a public place, let alone from their own property.
> ... they [the average person] do have control over whether or not they install a camera of their own.
That's true but let me show you why whether or not the average person installs a camera of their own doesn't matter. Take my neighbours for example; I have over 170 of them, identities, historical movements and current location estimates tracked and searchable from just the FOV of one web camera in my living room window. And that's from below average residential traffic.
That's 1 person out of 170, far from the average. They didn't need to install a camera to be part of the network. They were part of it long before they'd even heard of it.
You mention 1 person out of 170. Okay, I think I have more than 170 neighbors whose windows I pass in order to leave my neighborhood. I can't imagine even one choosing to install a camera. That's the point I'm making: Many/most people will not rationally weigh the costs vs benefits of this system. They will see only "more cameras, less privacy," and that's that.
BTW, there is no legal expectation of privacy, but that's a far cry from the average person's understanding. We live in a world in which people attach notes to Venmo payments spelling out their illicit activities, in large part because they seem to not realize just how public that information is.
The network you describes seems to rely on a regular series of periodic updates to track movement. I can't imagine uptake being high enough in most locales for that to work.
The full version has a lot of my own personal configurations in there so I haven't uploaded it yet. It also requires a good video card so it's expensive to run.
I'm still working on the lite version as you've probably seen. It comes down to trying to make the right tradeoffs when computational power is scarce. I'll be uploading the lite version's tandem Detection Server as a Docker file along with it.
Which one are you interested in?