A Proposal of N ew Visual Servoing Method with Gray-Scale Image and GA -Traeking Experiments to Swimming Fish-

摘要

We report a robot system with vision which follows free motion target by recognizing it in real time. We employ Genetic Algorithm (GA) and the gray-scale image termed here as raw-image. GA is used to perform the search for a target object in the raw-image by an exploration of the search space and exploitation of the best solutions of this search process. And GA is used in a way that we named I -step GA Evolution since for every generation， that is， every step of GA evolution， GA searching results which are position and orientation of recognized target object are output for the robot controller as position/orientation servoing command. This GA search is based on a concept of model fi1tering which is proposed in this report. In order to determine which object-model is the best， that is， which one provides the best recognition resu1ts criteria， three object-models， namely a frame model， a surface model， and a surface-strips model are examined in this research. From an analysis of the object-models sensitivity to position and orientation variations， the surface-strips model happened to provide more accurate and reliable recognition results. But the surface-strips model， when compared to the surface model from the view point of the convergence speed， has been found to have a slower convergence speed. Therefore the surface model can be seen in a certain way to be suitable for real time control. Our experimental servoing system with a hand-eye camera works effectively to track its eye position to a swimming fish in noisy environment with water plants. Furthermore， we improved the tracking performance by adding local searching mode around the target object in GA， which is inspired from gazing action of human.