Ball Detection Under Variety Lighting using Omnidirectional Camera

Abstract

Detecting ball is the first competence that should be owned by a robot soccer. In the case of wheeled-based robot soccer, the existence of ball is detected using a usb-based web camera combined with an omnidirectional mirror. Traditional wheeled-based robot soccer usually uses color thresholding technique to differentiate ball and other objects. However, the accuracy of thresholding technique frequently decreasing due to direct interaction between ball and lighting conditions. This condition causes the robot can not detect the ball correctly. In this paper, we apply an illumination invariant approach for our ball detection system. This approach utilizes an extended illumination model combined with Fuzzy Inference System (FIS) to control model's contrast ratio. The rule of FIS was optimized first using Genetic Algorithm (GA) so that the rule of Fuzzy roles as an illumination filter coefficients. Our experiment under severe lighting conditions shows our proposed technique outperforms the thresholding technique and very promising to be used in real applications.