Colour Pattern Recognition with Two-Dimensional Rotation and Scaling for Robotics Vision Using Normalized Cross-Correlation

This paper introduces a novel approach to recognise two dimensional (2D) colour pattern objects having different rotation and scaling. Different rotation and scaling of the desired objects can be shown in an input image; detection of these objects requires numerous iterations which is time consuming. In this paper, several candidate objects are detected first, and then their orientation and scaling on the input image are obtained. The reference image of the object is then altered to assume the rotation and scaling of the object of interest before the correlation method is performed. The desired object is then detected by NCC (normalised cross-correlation). Used in a pick-and-place sorting robot application, an object's unique elliptical parameters are determined. This approach significantly reduces the number of iterations required, by knowing and searching for objects' orientation characteristics. Statistical analysis is applied to reduce potential errors arising from ambiguity in the solution or when similar objects are 180 deg, or close to 180 deg rotated relative to each other. By adapting NCC threshold limits as statistically indicated, simulated results show improved object detection.