What I have in mind is something like techniques where visible edges, maybe along with estimated tangent vectors relative to an estimated view-angle, are used to extrapolate likely positions for invisible edges. E.g. if an object appears to be a rectangular solid -- based on its visible portion -- then we can expect the invisible half to be basically a mirror-image of the visible half, so we could infer the lines the camera would pick up were the edges not included by the rest of the solid. Likewise for something shaped like a cylinder if we imagine edges to be elliptical, so that once a proper eccentricity is calculated the occluded backside of edges could be extrapolated to "wrap around". Practically, I imagine these sorts of inferred edges could be drawn on the image representing edges that would appear in the absence of occlusion.
I'd like to know if there is standard terminology for the computational steps involved for this kind of analysis (estimating camera angles, tangent vectors to inferred curves in 3-space, etc.) and/or well-known algorithms -- analogous to Hough line detectors, for instance -- for this problem.
I would be interested in seeing OpenCV code if functions exist in the distro, but I haven't found them -- maybe I don't know the proper terms to search for?
