Comparison of Color Identification on Soccer Robot using Color Filtering, k-NN and Naive Bayes

Abstract

Accuracy level and computation time to identify color object are two important issues in designing vision of soccer robot. In this research three classification methods were used to solve those problems, i.e. Color Filtering, Naïve Bayes and k-Nearest Neighbor (k-NN). These methods were used for color-based image segmentation, object detection, center point coordinate and object distance measurement with scanning and tracking action. The result shows the average error of 3.6% by means of Naïve Bayes classification within distance estimation of 10 cm up to 360 cm. The fastest computation time was required for the object color identification using Color Filtering, i.e. 0.097 s. The average error and computation time resulted by k-NN method were 7.19% and 0.27 s, respectively.