自动机器人的生物视觉系统.pdf

Neurocomputing 38}40 (2001) 1385} 1391 A biologically inspired visual system for an autonomous robot Luis F. Lago-FernaH ndez*, Manuel A. SaH nchez-Montan eH s, Fernando Corbacho E.T.S. de IngeniernH a Informa tica, Universidad Auto noma de Madrid, Madrid 28049, Spain Abstract We have implemented an articial visual system that takes advantage of known properties of biological systems to achieve segmentation and recognition of simple images. The use of biologically plausible mechanisms makes the system inherit a series of features that are present in biological systems, such as #exibility, robustness and adaptability. The implementation of the model on an autonomous robot has proved its reliability and robustness in real environments and shows the relevance of this kind of approach. 2001 Elsevier Science B.V. All rights reserved. Keywords: Articial vision; Synchronization; Visual cortex; Unsupervised learning; Auton- omous agents 1. Introduction The success of biological agents in a great variety of environments is a clear indicator of possible directions to follow in the design of articial autonomous systems. The general principles of organization that seem to be crucial for the success of biological systems must be an essential reference for engineers. Here we describe the implementation of an articial visual system that is inspired by neural principles known to operate in the mammalian visual cortex, such as synchronization of neural responses and unsupervised learning. The model processes the visual stimuli in three di!erent stages (see Fig. 1). First a pre-processing of the image captured by a video camera is performed in order to generate a suitable input for the next processing steps. * Correspon