Sistema de localização de gestos utilizando um sistema multicâmeras

Abstract

Intelligent systems capable of understanding human motion can be used for several purposes. That is why there is a great effort to develop and include such systems in our daily lives. They can be used to provide, for instance, interfaces for human-machine interaction or even help human-human interaction, movement analisys for people rehabilitation, detection of suspicious activities and many other applications. In this context, this Master’s Thesis presents components of a gesture localization system, addressing practical aspects for its implementation. Gesture detection or localization means to determine the instants when a gesture starts and ends. The problem was tackled in two steps. The first is to obtain the three-dimensional location of humans, using just a calibrated multicamera system. The following step is responsible for detecting the execution of gestures or actions using human pose. For both steps, the method was validated using known databases, and calculating the accuracy and feasibility of using it in real time applications. The developed system may be used by other works that apply either human pose or gesture detection for their purposes. Finally, a discussion is made involving an overview of the system that shows how both developed methods can be used together