Constructing natural neighbor interpolation based grid DEM using CUDA

Abstract

Constructing digitial elevation model(DEM) from dense LiDAR points becomes increasingly important. Natural Neighbor Interpolation (NNI) is a popular approach to DEM construction from point datasets but is computationally intensive. In this study, we present a set of General Purpose computing Graphics Processing Unit(GPGPU) based algorithms that can significant speed up the process. Evaluating three real world LiDAR datasets each contains 6~7 million points shows that our CUDA based implementation on a NVIDIA GTX 480 GPU card is several times to nearly 2 orders faster than the current state-of-the-art NNI based DEM construction using graphics hardware acceleration.