Simple, Thread-Safe, Approximate Nearest Neighbor Algorithm
Michael Connor · 2007
This thesis describes the implementation of a fast, dynamic, approximate, nearestneighbor search algorithm that works well in fixed dimensions (d ≤ 5), based on sorting points in Morton (or z-) ordering. This algorithm scales well on multi-core/cpu shared memory systems, and can run on multiple processors simultaneously. The implementation is competitive with the best approximate nearest neighbor searching codes available on the web [1], especially for creating approximate k-nearest neighbor graphs of a point cloud. An extensive C++ library has been built implementing the research presented here. It can be found at: http://www.compgeom.com/∼stann.