Visual Analyzer of the Animat as a Semantic Basis of Robot Sensoric System

摘要

In the framework of study the implementation of social behavior models in group robotics systems, the question of the arrangement of animat visual system is of high importance. In this paper we propose the structure of the so-called visual analyzer, simulating the work of the perception system of the animat. The direction of modeling animated ants (artificial animals) has been chosen in the robot society design. The system requirements have been developed: the ability to recognize indiViduals, determine their number in the field of View, giVe information about the change of its quantity, and distinguish its poses. The system was implemented in C++ using technical vision and graph analysis libraries. The video stream passes through three consistent processing stages: selecting attributes, recognizing simple and complex objects, recognizing scenes. The method of recognizing objects is realized by decomposing their features, the algorithm for recognizing complex objects and scenes is based on the search for isomorphic subgraphs in the system of technical vision. The principal possibility of using such a system for recognizing models of animats of different kinds, scenes with participation, formation of simple kinoptical reactions is shown. A software and hardware complex designed for installation on a mobile robot is described. Models of animated ants were simplified for the experiments. This was done to test the applicability of the approach. After testing the system on the drawn abstractions, the system was tested on real images received from the camera. Further development options for the system are indicated.