| Abstrakt: |
Topographic Light Detection and Ranging (LiDAR) captures geometric information of the topography of a geographical region, often using airborne platforms. The research and practice of analysis of point clouds acquired using LiDAR is more recent in comparison to that of LiDAR imagery. Point clouds are unstructured datasets, where its geometric or structural classification labels the constituent points as belonging to line-, surface-, or point-type features. We focus on line-type features in the LiDAR point clouds of urban residential areas, which enables extraction of building outlines. We use a multiscale local geometric descriptor (LGD), computed using tensor voting and gradient energy tensor to enhance specific line-type features, e.g., gable roofs. Given that LGDs are positive-semidefinite second-order tensors, we propose a tensor-based data analytic workflow for extraction of boundaries in building roofs using the LGD. We use the tensor representation of the LGD to extract “tensorlines,” which are then postprocessed for extracting feature lines of the building roofs. Our proposed workflow provides the flexibility to the human-in-the-loop for exploration of point clouds for roof boundary tracing for selected buildings. We demonstrate the workflow for a two-plane gable roof. [ABSTRACT FROM AUTHOR] |
