I've also noted that red light cameras are overly simplistic. They simply take a snapshot, which may appear to mean one thing, but should run continuous video, then just trigger on certain events.
The video buffer could be very small (20 seconds or so) so each action would be lost after 20 seconds, unless a triggering event causes the previous 20 seconds to be saved for review. Review would often show that the person violating a red light was simply avoiding a conflict with someone else who was violating the law, in another way.
Actually they take a series of pictures. They don’t care if you were legit in being in the intersection, these are revenue generators not justice.