Registration of multi-platform point clouds using edge detection for rockfall monitoring

Remote sensing methods which produce point clouds, such as terrestrial laser scanning (TLS), terrestrial photogrammetry (TP), and small unmanned aerial systems (sUAS) have become an integral component of geotechnical monitoring programs. Applications such as rock-slope monitoring benefit from multi-platform datasets to be acquired in two or more different epochs. Accurate registration of these datasets in a common coordinate system is essential for detecting slope changes. Their registration often relies on initial feature-based alignment followed by fine alignment with the iterative closest point (ICP) algorithm. When practical, ground control points (GCPs) and other surveying targets with well-defined coordinates are used. However, establishing such GCPs on rock surfaces can be difficult, expensive and dangerous. In addition, GCPs and targets can be lost or destroyed with time and re-establishing them is difficult. This paper develops an automated registration algorithm based on edge detection that can register multi-platform and multi-epoch point clouds. Rock-surface edges are expected to remain largely the same and be captured in point clouds collected in two different epochs. For edge detection, we use α-molecules that offer a unified framework of most multi-scale transforms that can be adapted to any rock-surface. Then the algorithm identifies edge correspondences based on the discrete Fréchet distance. From the corresponding edges we derive matching points between datasets. Transformation parameters are then derived through Procrustes analysis. Using real and simulated scenarios, we demonstrate the utility and performance of the proposed algorithm. For example, sUAS scenarios with 0 and 1 GCPs show that initial root mean square error (RMSE) values of a few decimeters drop to a few centimeters. Scenarios with simulated translations, rotations, and scale showed that the developed algorithm registers multi-platform point clouds with mm differences from their original RMSE values. Results demonstrate that the algorithm can successfully register multi-platform point clouds and support rockfall monitoring.