YUV based automatic colour calibration for NAO robots

Abstract

A challenge in real time application of NAO soccer robots is in colour calibration. Many tasks such as localisation and goal detection rely on robustness of colour calibration. In this paper a robust and accurate YUV colour space based automatic colour calibration technique is proposed. First the specific set of frames from the NAO's camera has been analysed in order to define average values for desired colour classes, namely orange, white, green and purple. Then those average values are corrected by a luminance analysis of a new frame and are passed to the K-means clustering algorithm as a set of initial means. Apart from those 4 values, set of initial means of the K-means algorithm also contains 16 values that are calculated in the following manner: the frame currently being processed is divided into 4 by 4 grid and average value from every grid will serve as an initial mean for K-means clustering. After the K-means clustering is applied to the frame so that colours of a similar type are combined into clusters. Final step of the proposed technique is the cluster classification, which is performed by measuring the distance from cluster centroids to the previously calculated average values of desired colour classes corrected by luminance analysis. The proposed colour calibration technique has been tested on white goal detection.