Ball detection algorithm for robot soccer based on contour and gradient hough circle transform

摘要

Kontes Robot Indonesia (KRI) changes their rule every year. This paper proposes a new ball detection method in response to the changes of the rules on Kontes Robot Sepak Bola Indonesia (KRSBI) division for KRI 2017. For the humanoid division the ball is changed from the orange tennis ball in previous year's competition to a white, size 1 FIFA soccer ball. For the wheeled division an orange size 4 FIFA soccer ball will be used. The algorithm proposed in this paper is based on the contour and Gradient Hough Circle Transform (GCHT) which is capable of detecting balls of any color. The contour based will be primarily used to detect the ball and the GCHT will be used for the second chance if the robot fails. Experiments show that the proposed algorithm can achieve ball detection accuracy of 100% (orange ball) and up to 96% (white ball) fixed-placement experiment. In random-placement experiment, the proposed algorithm achieved an accuracy of 93.3% (orange ball) and 96.6% (white ball). Furthermore, the processing time observed in the experiments is quite fast, averaging at 37.76 FPS.