Indexing of LiDAR Point Clouds during Capture

Mobile LiDAR scanners produce large amounts of raw point data. This data will often not fit into main memory. Visualizing large LiDAR point clouds therefore requires the use of out-of-core methods and multiple levels of detail. Spatial index structures that are used to enable these requirements are usually created in a batch processing step after the point cloud is acquired. This thesis explores, if and how such index structures could be created already during the capturing of the point cloud. This simplifies the point cloud creation workflow by eliminating the need for the extra processing step. Also, the index structure can be already used during capture, enabling a high quality live visualisation of the recorded point cloud. Two possible index structures for indexing of LiDAR point clouds during capture, the octree index and the sensor position index, are introduced and implemented. In addition, a point cloud viewer that is based on the index structures, was implemented as a proof of concept. Several optimisations to the indexing process allow to index points fast enough to keep up with typical point rates of current LiDAR scanners. These optimisations include an LRU cache for avoiding excessive disk accesses, a parallel implementation of the indexing process, and for the octree index optimized scheduling of the node processing order. The evaluation shows, that the octree index fulfils all requirements for indexing during capture. The sensor position index has disadvantages in query execution and parallelizability, but can index more points per second.