I like that kind of problem.
One idea is a rough LOS check. Basically repeatedly pick a random point inside the collision poly a do a LOS with that. Would fail more with smaller loops if the object is obstructed but it may give a true eventually. Could be a useful approximation of vision. Anyways the main sub problem to solve would be to get a random point in a polygon.
A second idea is to utilize raycasts. Either a uniform spread of rays in roughly the direction of the object or use a bit of random. If any of the rays hit, there is LOS. I think this one is the most doable. Not sure if the ray cast feature in c3 can be used for this but making a ray cast with events would work.
Another idea if to calculate the view polygon from one point and check to see if that view poly overlaps an objects collision poly. May be tricky to calculate though. One approach would be to do what shadow casting does and clip away from the collision poly the shadow bits. This would give an absolute is in in Los or not check.
Anyways just some ideas, each with their own set of sub problems to solve. Maybe some are useful.