A methodology for pattern matching of complex objects

Abstract A complex object is an entity made up of movable subparts. This paper presents a method to recognize and determine the position and orientation of an observed complex object with respect to a database of known objects. Each database complex object is described by a special type of relational graph, which guides a localized matching algorithm to determine the position and orientation of each complex object subpart. The feasibility of the assignments based on the kinematics of the subparts is then determined. Global assignments are made based on the feasibility of the local assignments.