Object Detection Using Deep Learning for a Humanoid Soccer Robot

Abstract

In this paper, we present a real-time object detection system for a humanoid soccer robot. We employed YOLO, a deep-learning-based object detector, and trained it to detect soccer balls and goal posts from an image. For efficient image annotation, we developed a GUI tool equipped with a semi-automatic ball detector. We evaluated the ball detection performance of the system and observed superior performance over an existing non-deep method. Furthermore, we implemented the object detection system on our humanoid soccer robots and participated in RoboCup 2017 competition. The problems we faced during the competitions and how we overcame the problems are detailed in the paper.