Combinatorial Characterization of Perspective Projections from Polyhedral Object Scenes

Model-based computer vision systems which recognize objects in single grayscale images require the matching of stored object models and the image data resulting from the perspective projection. If objects may be located arbitrarily in relation to each other occlusions can occur, thereby creating different line configurations in the projected image with different viewing directions. For a collection of polyhedral objects containing n vertices, there are of the order O(n3) different general views and O(n2) degenerate views and the algoritmic complexity for constructing a view list is O(n3), where n is the sum of polyhedral bounding faces part of the assembly convex hull and planes arising from visual interaction of polyhedral parts inside the convex hull.