Reconhecimento de peças do jogo Progster utilizando aprendizado profundo

Abstract

This work proposes a new approach in one of the processes of the game Progster, making the use of Deep Learning to the classification of the regions of interest of its board. The objective of this work is to create Deep Learning models that classify the regions of interest of the board and can be used in an application that discards the previously proposed electronic board, making the game more accessible. For this, we used the techniques of Computational Vision in the stage of creation and data collection of regions of interest of the board. From the data collected, Deep Learning techniques were applied to classify each region of interest of the board. The results showed that in the best experimental scenario the accuracy of the classification reached 99.37 %, evidencing the current effectiveness of the Deep Learning methods in image classification problems.