Concepts and Challenges for 4D Point Clouds as a Foundation of Conscious, Smart City Systems.

Point clouds represent the as-is geometry of indoor and outdoor environments by sets of 3D points. They allow for constructing 3D models of objects, sites, cities, and landscapes and, hence, form the base data for almost any conscious, smart city system and application. For implementing such systems, we need a spatio-temporal data structure that enables efficient storage and access to 4D point clouds. In particular, the data structure should allow continuous updates, change tracking, and support for spatial and spatio-temporal analysis. This paper discusses challenges and approaches for a 4D point cloud data structure. In particular, the challenges arise from repeated scanning of environments in terms of sparsity, data redundancy, and geometric blurring of the corresponding point clouds. We outline a scheme for incremental storage of 4D point clouds via signed distance fields using a sparse, voxel-based representation. To efficiently implement analysis operations, we discuss how the data structure supports access based on both spatial and temporal criteria. In particular, we outline how machine learning-based interpretations used to classify point clouds and derive object-based information can work with the data structure.