Deteksi Objek Robot Sepakbola secara Real Time dengan Mobilenet Single Shot Detector (SSD)

Abstract

Football is a popular sport in the world. Along with the development of technology, in 1992 the idea of robots playing football emerged from Professor Alan Mackworth (University of British Columbia, Canada) in his paper entitled "On Seeing Robots". This idea was finally executed in 1997 with the participation of 40 teams and more than 5,000 spectators. In a robot football game, there should be no human intervention (full autonomous system). The robot uses computer vision to detect important elements such as the ball, line, goal, and robot opponents. In the game, the robot is equipped with a single board computer. This research utilises the deep learning Single Shot Detector (SSD) algorithm with MobileNet as the backbone which produces 90% accuracy with Intersection of Union (IoU) threshold ≥ 0.5. and produces 2 frames per second when deployed on the Raspberry Pi platform.