I'm perfectly willing to entertain answers like, "You shouldn't, do it like this instead," so I'll explain my ultimate goal.
I've been working on a model in Blender, including materials, and I'd like to prepare textures for use in a different, much more primitive engine. Because I've been building for Blender appearance, I would like to use renders from orthographic cameras (starting with 6, but we'll see) to build a texture up on an unwrapped UV.
My problem is concavity. Some areas of the model will be occluded from any given camera-- from all cameras. I can manage that by separating these faces out, but I'll need to know which ones. The ideal way for me to know would be to have a texture on my unwrapped UV that represented whether a given pixel was visible to the camera (or any proxy for it), in a black-or-white fashion, no grayscale, so that I could feed that into a nodes setup to integrate all of my renders into a final image.
Is there some way that I can create a material that returns white if and only if it is visible from a given position with an orthographic perspective? Obviously, I can do this just fine for the render-- everything that renders is visible and everything that doesn't, isn't-- but I need it mapped to my model's unwrapped UV coordinates, like a bake.