I think #3 is going to be the sticking point. Pretty much by definition, visual separation means you're able to get quite close to other traffic and avoid them at the last moment (say, on a direct collision path until five seconds out), and this is 'acceptable'. So you can't give any kind of serious alarm ten/twenty/thirty seconds out, because 99% of the time it'll be 'totally under control' and an annoying false alarm.
To be effective, any collision warning/avoidance system needs to give you enough notice to be able to avoid traffic, meaning it has to activate before your last chance to safely avoid them. To not give a huge number of false alarms, it has to activate after you would have needed to act to normally avoid them.
Visual simply doesn't give a gap between those those two times, so a collision avoidance system is not really practical. Enhancing awareness is all it can do.
Anyone who's sat with a learner driver through roundabouts/gap selection has doubtless experienced the trouble you have doing this.