Digital Image Processing In Robot Vision

ii Abstract This report documents the modifications made to the frame processing module of the NUbot vision system for improvement in robot soccer competition and particularly in the variable lighting challenge. The frame processing module involves the classification of images with respect to a look up table (LUT), a new system of secondary colour classification for blue robots that uses edge detection, blob formation and blob filtering. Soft colour classification was used as the basis of improvement in frame processing. This involved several specific modifications to the previous system. Additional soft colours were integrated into colour classification to assist the differentiation between field objects of similar shades. A secondary classification method that uses edge detection was implemented to account for the unique edge-colour characteristics of blue uniformed robots. This method allows the classification of blue robots; this has not been achieved effectively using standard classification methods. Modifications to the blob formation method were required to account for the soft colour classification system. The modified blobs were then filtered to extract useful data, which is interpreted by object recognition. The LUT system used in the classification of images was modified to increase accuracy and robustness of object data. The modification involved the implementation of a 7*7*7 bit LUT system. Another improvement of the system involved the modification and implementation of a Support Vector Machine (SVM) to partially automate and hence optimise LUT generation. The modifications to vision have improved accuracy in object data and robustness in colour classification. Additionally, these modifications are effective under wider range of lighting conditions than previously. • Implement a secondary colour classification system that uses edge detection to classify blue robots • LUT files for use in colour classification for test situations and competition.