Machine vision for multi agent autonomous robots

Abstract

Dynamic, adverse, and uncertain. Robotics Soccer has became a very challenging task in robotics field. This works presents a new approach for machine vision for multi agent autonomous robots. The first part involves improvement of accuracy of color segmentation. A tool is developed to help developer to define threshold map accurately. The second part involves an engine which parses runs based on the pre-defined grammar so that the semantics of the runs is determined before object recognition stage starts. The abstracts the operation and enables object recognition engine currently used can be re-use. A challenge from RoboCup 2003 is selected and algorithms are implemented on this challenge. The third part presents a ways for multi agent system to archive object localization without prior knowledge of the size of the target object. This is done by finding intersection of line of sight of agents to the target object. The related idea will be discussed and results of demonstration are shown as well.