RoboCup Soccer Ball Depth Detection using Convolutional Neural Networks

Abstract

RoboCup, a soccer competition played by autonomous robots, raises a variety of hard problems in different fields such as Artificial Intelligence and Dynamics. In particular, vision is the essential input to take actions based on this highly dynamic environment and, although traditional techniques might be cheaper for object detection, new rules such as multicolored balls require more robust detection methods and allow a significant computing power improvement. Given these recent changes, this work presents a solution for ball localization using convolutional neural networks (CNNs), a largely applied machine learning technique holding the state of the art technology for several imagery problems. To perform the object detection, we train the CNN with robot’s camera images as input and the ball’s coordinates relative to the robot as output.

5 Figures and Tables

Cite this paper

@inproceedings{CAMPINAS2017RoboCupSB, title={RoboCup Soccer Ball Depth Detection using Convolutional Neural Networks}, author={ESTADUAL DE CAMPINAS and G. Milit{\~a}o and Esther Luna Colombini and Gabriel Milit{\~a}o}, year={2017} }