I don’t really have anything that goes beyond basic experiments. Many aspects of the goals had not been clear. And a sandcastle like the following may not really provide any substantial help…
It just defines a “HUD Canvas” (in the HTML part), and draws some arbitrary line into that. The computation of that line is marked with
// XXX THIS DOES NOT MAKE SENSE!!!
because I removed all my experiments for detecting the actual horizon line, and replaced it with some “dummy” line computation.
One can still see what the goal was…
and the orientation of that red line is actually “correct”, because it only depends on the roll angle. But implementing that properly would require more time…