A ConvNet-Based Approach Applied to the Gesticulation Control of a Social Robot

Abstract

This document presents the enforcement of a facial gesture recognition system through applying a Convolutional Neural Network (CNN) algorithm for gesticulation of an interactive social robot with humanoid appearance, which was designed in order to accomplish the thematic proposed. Furthermore, it is incorporated into it a hearing communication system for Human-Robot interaction throughout the use of visemes, by coordinating the robots mouth movement with the processed audio of the text converted to the robot’s voice (text to speech). The precision achieved by the CNN incorporated in the social-interactive robot is 61%, while the synchronization system between the robot’s mouth and the robot’s audio-voice differs from 0.1 s. In this way, it is pretended to endow mechanisms social robots for a naturally interaction with people, thus facilitating the appliance of them in the fields of childrens teaching-learning, medical therapies and as entertainment means.