I'd assume the los behavior shoots raycasts at all imagepoints to attempt to make sure that even if a small part of the object peeks out, it triggers the los. With such a small cone you'll shoot through these objects if it doesn't happen to hit an imagepoint + I think the objects being a obstacle themselves will potentially obscure image points on the backside that otherwise would trigger the los.
Meaning small gaps can also fail to register los even if you set the cone bigger if no imagepoint happens to be visible behind the gap.
I guess the way to mitigate that would be to cast a bunch of raycasts along the lines defined by the image points, rather than only the imagepoints themselves, with the tradeoff being that it's more expensive to do. Although I've been told raycasts are super cheap. OR you increase the amount of imagepoints but I don't think that's the smart way.