“…For polygonal objects of fixed shape but variable size, matching is normally attempted by assembling collections of primitive features such as edges / lines as in [2,3], or corners / geometric invariants as in [1,9,10]. As opposed to other methods in which polygon are detected by making a single attempt to match models to images, the efficient particle filtering approach in [7] combines many attempts, based on sequential Monte Carlo sampling, to track a single polygon boundary.…”