Keypoints of the current image and the map image are detected and described using the SIFT algorithm. The keypoint sets are further refined using Lowe’s ratio test \cite{lowe1999object}. This is followed by a matching process, that identifies corresponding keypoints between both images. The matching uses a ’brute-force’ matching scheme and every keypoint is compared to every other keypoint. These matches allow for finding a homography between both images. For determining the \(x,y\)-position of the current image, the center of it is projected on the reference image using the homography matrix. The pixel position of the center in the reference image can be used to determine the real world position by transforming the pixel coordinates to real-world coordinates, based on the scale factors \(C_{x}\) and \(C_{y}\), with \(C_{x}=\frac{width(R)}{width(I)}\) and \(C_{y}=\frac{height(R)}{height(I)}\), where \(W\) is the real-world representation and \(I\) the digital pixel image. This yields a dataset of images, labeled with \(x,y\) coordinates and the number of matches. This process already introduces noise into the dataset, depending on the quality of the keypoint matches.