Spatial data storage and processing strategies for urban laser scanning

Today, laser scanning technology offers unprecedented data for a wide range of crucial analysis of the complex urban environment. However, the significant data burden and the data complexity have remained significant impediment to effective data exploration. The latest commercial laser scanning instrument is capable of conducting in excess of a million spatial and temporal measurements every second. Since laser scanning projects are typically deployed over large geographical areas, this results in large and complex datasets that require sophisticated data modelling and indexing strategies to ensure efficient data accessibility. Additionally, such voluminous laser scanning data frequently requires strategic processing strategies to extract semantic information so that they can serve specific demands in various application domains.Given the large and rapidly growing quantities of laser scanning data being produced, many existing laser scanning data handling solutions are quickly losing their viability, thereby requiring new improvements in the field. The first part of this thesis proposes two innovative LiDAR data handling solutions aimed at scalability, data retrieval speed and advanced functionalities. They include a system for discrete point data, called UMG_PC (Urban Modelling Group - Point Cloud), and another data system for full waveform LiDAR data, called UMG_FW (Urban Modelling Group - Full Waveform). Both LiDAR data handling systems are based on a novel hybrid spatial indexing strategy, which combines a two-dimensional indexing structure at a global level and multiple, in-memory three-dimensional indices at a local level so that the entire index adapts to the typical spatial distribution of urban LiDAR datasets. The point cloud database system is highly scalable while simultaneously offering better data retrieval speed compared to the traditional one level indexing solution. In addition, it is also capable of supporting advanced queries, including incremental nearest neighbour search and planar segment selection, which was not previously available in a spatial database. Furthermore, the full waveform data management system, with its capability of supporting spatial and spatial-temporal pulse data retrievals, represents a breakthrough in storage and indexing of full waveform LiDAR data.The second part of the thesis devises several improvements to two challenging topics in LiDAR processing towards automatic urban modelling. The first is the integration of an octree spatial structure into a region growing segmentation algorithm, which vastly accelerates the data processing speed without significantly compromising the output accuracy. In that work, the octree plays multiple roles: (1) data simplification, (2) neighbourhood identification, and (3) data indexing. Additionally, a workflow is proposed for detection of complex urban road networks from a dense aerial point cloud fused with high-resolution aerial imagery data. This includes a novel algorithm that exploits the high variance of slope and height of the point data in the direction orthogonal to the road boundaries. That final study demonstrates an end-to-end LiDAR processing workflow making use of data residing in a database. That study reveals emerging opportunities for detection of small objects from airborne laser data.