A New Method for Change Detection in Stone and Concrete Structures with Digital Photogrammetry

Abstract

Point clouds, generated by Photogrammetry and LiDAR, allow us to detect far more complex deformation processes, with greater speed and accuracy, than previous surveying techniques. The accuracy of any monitoring is determined, in part, by the method used to align the two epochs in the same space. Surveying equipment can be used to stake-out ground control points visible in either the laser scan or the photogrammetry and so provide absolute positioning for each epoch. As an alternative, 3D shape-matching algorithms like RANSAC and Iterative Closest Point (ICP) can be used downstream to align the two epochs based only on the invariant features visible in the dense clouds. Shape-matching is, however, limited in accuracy because a) operator proficiency in guiding the algorithms leaves room for error, and b) the point clouds used for alignment are usually insufficiently dense to guarantee sub-centimeter change detection. Proposed is a new method of alignment between monitoring epochs for photogrammetry. Photogrammetry has the ability to match features between images down to the accuracy of 0.15 pixels, and provides us with a robust statistical model to predict both accuracy and error. Using invariant features in photographs shared between monitoring epochs, we can use the algorithms of Normalized Cross Correlation and Least-Squares Matching to very accurately "pin" one monitoring epoch to another. In a series of simple experiments we demonstrate that sub-centimeter change detection is easily accomplished on highly textured surfaces such as stone and concrete. This has applications in Archaeology, Architecture, and Civil Engineering surveys.