Yeah this is a tricky one. Modern gold standards are to use cameras to recognize pedestrians and cyclists to give them intersection priority. This is a big benefit if designed properly but video data this way poses a huge privacy/surveillance risk.

I suppose maybe there would be a way to process the footage locally and delete it as soon as it's no longer needed but that's going to require a lot of oversight from the community to confirm.

Stabby, I just wanted to say, I saw your original comment, and was going to side with you - I don't think it's feasible to avoid surveillance. I also wanted to say that the way this conversation touches on all the points I thought about 'drafting' my comment in my head, so it was really great to read.

While you can have reasonably dumb sensors (like motion lights in our house, but y'know in a city application), anything beyond that gets really challenging to use without compromising privacy, or becomes a tremendous pain the ass if we give a shit about privacy. Having a bajillion sensors all not talking to each other, or talking to each other through numerous layers creates a data management nightmare. it's why you don't get a bunch of technical experts to write several different parts of a report, for instance. You need one main author to go through and make it all sound the same and the parts to speak to each other. Same goes with data. Because of that need, you open yourself up to all that data/surveillance being compromised by 3rd parties or abused by governments, as you point out.