Knowledge Based Shape from Shading

摘要

Edge-based matching can differentiate many objects, however, it is inadequate for discriminating objects with a similar wire-frame. This chapter presents a method for extracting the shape of an object from a single image when knowledge about the object is available. The main application in our architecture is for shape-based verification of edge-based matches, where shape derived from the image will be checked against the shape for the candidate model view of the object. A second application is where a robot needs to recover object shape when it has only a wire-frame model,1 and it has to move close to the object due to obstacles or task requirements. Navigating entirely on the basis of the wire-frame may lead to a collision with a convex curved surface, even though the robot is clear of the object as modelled by the wire-frame, as shown in Figure 6.1.