Ball Detection using Local Binary Pattern in Middle Size Robot Soccer (ERSOW)

Abstract

Detecting ball is the first competence that should be owned by robot soccer. In the case of wheeled-based robot soccer, the existence of a ball is commonly detected by using an omnidirectional camera and a monocular camera. Traditional wheeled-based robot soccer usually uses color thresholding technique to differentiate ball and other objects. However, the accuracy of the thresholding technique frequently decreasing due to the changing of lighting conditions. This condition causes the robot can't detect the ball correctly. In this paper, we address that problem by applying a Local Binary Pattern (LBP) based ball detection system. Our experimental results show the performance of our system has percentages of success of 96.97% in bright lighting conditions, 80.69% insufficient lighting conditions, and 98.74% in dimmed lighting conditions when the ball moves slowly. Whereas the percentage of successful detection becomes 70.74% during bright lighting conditions, 60.42% for sufficient lighting conditions, and 83.79% for minimum lighting conditions when the ball moves quickly. LBP-based ball detection has 92.4% accuracy in measuring the distance between the ball and the robot.