Scene Analysis For Navigation Tasks And Robotics Using A Harmony Theory Network

Abstract

This chapter describes a parallel implementation of a novel scene analysis system. The implementation is based on a Harmony Theory network, since Harmony Theory has been proved to be very competent at tasks requiring constraint propagation. The network (a) receives a thresholded projection of the viewed image (line-drawing) as its input, (b) utilises a novel system of l abelling schemes (precompiled information) for the construction of its upper layer and (c) outputs the decisions concerning the c haracterisation s of all the lines in the line-drawing (result of the network-settling ); these decisions constitute the 3-D description of the viewed scene. The rival objectives of transparency and economy in the construction are the focal points of this network. It is thus ensured that the implementation settles accurately and fast, in order to be appropriate for real applications.