Resumen
This study proposes an enhanced algorithm that outperforms the methods developed by the author?s earlier contributions for the recognition of railroad assets from LiDAR point clouds. The algorithm is improved by: (1) making it applicable to railroads with any slope; (2) employing Eigen decomposition for the rail seed point selection that makes it independent of the rails? dimensions; and (3) developing a computationally efficient fully data-driven method (simultaneous identification of rail tracks and contact cables) that is able to process poorly sampled datasets with complicated configurations. The upgraded algorithm is applied to two datasets with quite different point sampling and complexity. First dataset is scanned by a terrestrial system and contains three million points covering 630 m of an inter-city railroad corridor. It presents a simple configuration with nonintersecting straight rail tracks and cables. Second dataset includes 80 m of a complex urban railroad environment comprising curved and merging rail tracks and intersecting cables. It is scanned from an airborne platform and contains 165,000 points. The results indicate that all objects of interest are identified and the average recognition precision and accuracy of both datasets at the point cloud level are greater than 95%.