Human Body Recognition using Repro- ductive Perception Paradigm

An important issue in human robot interaction is for the robot to recognize the human agent. The problem has infinitely many application, for example surveillance [1], museums guide robots [2], car driving application [3] and many more [4] and also many problem specific solutions [5] . Apart from other application though, in rescue operation the task is even more important [5] because human agents can not replace robot agents in hazardous environments and therefore the autonomous machine has to be able to reliably and quickly recognize possible victims. Unfortunately, in most cases, rescue missions makes impossible to use problem specific solution because of the variation of the environment, human posture or high degree of occlusion. In this paper we propose a solution based on camera information that should behave relatively well unconditioned by the environment, human position or degree of occlusion. The idea is to use a combination of reproductive perception paradigm [6] [7] with standard methods, which basically means that the human body is detected by comparing the input data with generated pseudo data containing different body poses. The input data is partially processed to reduce the search space while pseudo data is generated using evolutionary algorithms to locate possible solutions. The approach proves to offer several advantages, it is robust against background changes, human posture variation and can handle occlusion. The results obtained showed that the algorithm offers a good trade off between speed and reliability and that further investigations, optimizations and generalization of the ideas definitely makes sense.