I mean, there are all kinds of ways to bound uncertainty. You’re right, you could film another camera, but that could be addressed separately. Digital signing would only prove that a video came directly from a particular camera’s sensors. To assert the veracity of a video, you’d then have to prove that particular camera was at the scene, and the light coming into the lens was authentic. Most court footage is incidental, not intentional, so the sort of doctoring you describe is unlikely in a majority of cases. Where it is likely, different measures could be taken to bound uncertainty.

Separately, generating false images or videos for libel purposes would become more difficult; if a camera that has never been owned by a person is used to produce a signed video that is pornographic in nature, the authenticity of that video is immediately suspect. It would take a much more sophisticated attack to produce a mathematically convincing fake video.

if you are worried about somebody actively faking footage, and are willing to go to such drastic DRM measures to prevent it, you need to explain how it is actually going to solve the problem, not just handwave.

I can imagine watermarking at the light sensor level. It would need some very special provisions for compression, though.

Problem is, this way journalists won't have anonymous sources anymore and you will be able to track any citizen filming power abuses.

Well, but then you'd have to trust the light sensor manufacturer to provide secure keys. Remember when the NSA hacked into Gemalto and had a chance to steal its SIM card keys?

However, that's still interesting; isn't it possible that each light sensor has certain watermark-like characteristics arising from the manufacturing process?

Yes, fingerprinting of digital cameras has been a thing for at least a decade [https://www.eurekalert.org/pub_releases/2006-04/bu-bur041806...]