Welp, if you’re really concerned with granularity such that bone occlusion is not enough, might I suggest vertex occlusion? You’ll need C++ for this, but you can get skeletal mesh vertex locations from the render buffer and transform them to the photo subject’s actor location then send the traces. Those that are not occluded are divided by the total number of vertices to give you a visibility percentage.
Dot product works for height differential, “Get Horizontal Dot Product To” which ignores Z.
For distance concerns, you only need to get the distance between the start and end vectors for the trace and determine a multiplier for the distance to each bone. For example, if a bone is more than 500 units away, its multiplier is 0.1 while 100 units away gives a multiplier of 1.0. Then average all the multipliers.
FOV changes seems like a design decision that’s in your court rather than a technical consideration, but when taking the actual ‘photo’ you could enforce a certain FOV for scoring before taking the picture then reverting back to the player’s preferred FOV for gameplay.