Recognition of parametrised models from 3D data

This paper describes work done as part of the Oxford AGV project towards recognition of classes of objects encountered in a factory environment. The authors address the problem of recognising an object from range-data observations as an instance of a parametric model class, and determining the values of the class parameters for that instance of the model, and the pose of the object. They represent an object class as a map from an underlying shape, a set of parameters, and some constraints on these parameters, to an instance of the class. A search of the interpretation tree is combined with a constraint network to determine the legal interpretations and parameter values using observations on an instance of the class. The authors demonstrate the feasibility of this approach using polyhedral models and simple range-image features (position, surface normal observations)