Desenvolvimento de uma interface humano-robô utilizando visão computacional e sistemas a eventos discretos

Abstract

This work presents a computer vision human-robot interface based on gesture recognition. A grammar composed of strings of simple gestures is defined so that an isolated gesture has no meaning to the system, and only a sequence of gestures performed in the correct order, according to the grammar, will issue a command to the robot. Each of these words, made of gestures, can be associated to a command, which will be issued to the robot everytime it's recognized. Through the use of this interface, a human operator can control a robot in a natural and intuitive way.Since only simple gestures are considered, the computer vision system is based on qualitative techniques that exhibits robustness properties. The output of the computer vision system is passed on to a stochastic discrete event system which is responsible for the commands recognition. Continuous gesture recognition is performed and the model implicitly identifies the beginning and the end of each gesture. Two types of discrete event systems were employed: Markov chains and Hidden Markov Models (HMM). The models were built automatically for both of these types.Experimental results shows that the proposed methodology yields robust recognition with low ocurrence of false positives even in complex backgrounds and with the operator moving. The HMMs outperformed the Markov chains in the proposed methodology context. The commands must be correctly chosen to avoid serious degradation of the performance as the number of considered commands increase.