NEURAL CLUSTERING OF CORRESPONDENCES FOR VISUAL POSE ESTIMATION

This paper is concerned with the problem of visual pose estimation, which entails, for example, the estimation of object translations. It adopts a correspondence based approach in general, and in particular, looks into a neural network implementation of the approach. The objective of the paper is to demonstrate how the approach can be learnt via the unsupervised clustering of correspondences into clusters representing different poses. Purely local (i.e. Hebbian) mechanisms were adopted in order to ensure not only the practical value of the learning algorithm but also its biological relevance. The results of the experiments here reported show that the learning strategy adopted allows for the successful unsupervised clustering of correspondences, even when the environment puts forth several difficult challenges, such as scarce or correlated features.