3D object recognition with a specialized mixtures of experts architecture

The aim of the AXON/sup 2/ project is the development of an object recognition system capable of recognizing isolated 3D objects from arbitrary views. Classification is often based on a single feature extracted from the original image. Here, we present an architecture adapted from the mixtures of experts algorithm which uses multiple neural networks to integrate different features. During training each neural network specializes in a subset of objects or object views appropriate to the properties of the corresponding feature space. In recognition mode the system dynamically chooses the most relevant features and combines them with maximum efficiency. The remaining less relevant features are not computed and therefore do not decelerate the recognition process. Thus, the algorithm is well suited for real-time applications.