An approach of a determination model about autonomous action using the depth cue from T-junctions

摘要

In this paper, we propose a model that can advance toward the direction determined by the information about T-junctions existing in the image data. T-junctions appear near the edge of an object when it overlaps the other objects or background and human recognizes the depth using this information about T-junctions. To detect an edge, we employ signal processing layers arranged by the nonlinear oscillator and find out the information about T-junctions. The present model can avoid an object under monocular view and low resolution pictures (32 /spl times/32 pixel). We construct an autonomous car, which moves, in a sense, unconsciously and carried out running experiments to examine the validity of this model.