O. Rusch, C. Ruwwe, and U. Zölzer (Germany)
Feature Matching, Correlation, RANSAC, ICP.
Several algorithms for matching feature points from dif ferent images have been proposed in the past. We give a brief overview of different matching techniques and show their application for maritime ship images. Due to the dif ficult image content in ship images, most standard algo rithms that are based on correlation or color similarity tend to fail. An extension for the widely used similarity con straint is given, but results on ship images are not usable for subsequent reconstruction purposes. Only a geomet ric matching approach like Iterative Closest Point is able to produce reasonable results. We will show that the latter is suitable for feature matching in maritime ship images. At least the projective transformation between the two images can be undone so that other content-based matching algo rithms will lead to more correct and robust results. Sub sequently applying RANSAC further improves the overall matching quality for three-dimensional reconstruction.
Important Links:
Go Back
