Landscapes, urban regions, cities, and infrastructure networks are captured with LiDAR or image-based remote sensing technologies in regular intervals. The resulting 3D point clouds represent a digital snapshot of the reality and are used for a growing number of applications. Multi-temporal point clouds, i.e., 4D point clouds, result from scans at different points in time and open up new ways to detect changes and establish workflows for updating and maintaining existing geodata such as 3D city models, terrain models, infrastructure models, and vegetation models (e.g., tree cadastre).
In this talk, we present techniques to manage, process, and analyze large-scale 4D point clouds. The input are point clouds from different acquisition campaigns without further information. A modular processing pipeline is used to analyze the surface structure, classify the data, and detect temporal changes fully automatically. Parallel GPU-based processing schemes and out-of-core strategies are used to reduce processing times significantly. Hence, 4D point clouds from entire cities and landscapes with billions of points can be handled.
Point clouds with different densities and characteristics (e.g., resulting from LiDAR and image-matching) from aerial and mobile mapping scans of different cities are used as case studies. Applications such as updating 3D city models, building up tree cadastres, and evaluating digital surface models are presented for datasets with up to 80 billion points.
In addition, we present visualization and interaction techniques that allow users to interactively explore, inspect, and analyze arbitrary large and dense point clouds in the web. Our web-based rendering system integrates additional data layers of point clouds and applies specialized filtering and highlighting techniques to facilitate the recognition of objects, semantics, and temporal changes within point cloud depictions.
The results show that the approach opens new ways to manage, process, analyze, and distribute large-scale, dense, and time-variant point clouds efficiently as required by a growing number of applications and systems.