Analysis of very large voxel datasets

The paper addresses the problem that large voxel models of detailed scenes in urban environments do not fit inside the (physical or virtual) memory of standard, or even high-end, workstations. It presents an investigation into out-of-core processing options and cloud solutions, while looking in particular for combinations of both. As a result, an octree-based datastructure is proposed. It is implemented in an already available key–value data store system. This solution has the advantage that existing in-core software can be quite easily converted into an out-of-core version. The resulting programs can still be executed on standard workstations with reasonable performance, while the expectation is that they can also be implemented in high-performance distributed multi-processor environments. •Analysis of volumetric voxel datasets too large for workstation memory.•Efficient out-of-core octree handling in a key–value data store.•Squared, instead of cubed, time and space complexity vs. resolution.•Comprehensive suite of generic, optimized, voxel analysis tools.•Allows for better performance with distributed NoSQL databases.