Tilewise Accumulated Cost Surface Computation with Graphics Processing Units

Abstract

Accumulated cost surfaces are used in a variety of fields that employ spatial analysis. Several algorithms have been suggested in the past for solving them efficiently or with minimal errors. Meanwhile, a new wave on the technological frontier has brought about general-purpose computing on GPUs. In this article, we describe how accumulated cost surfaces can be solved with CUDA. To verify the performance of our solution, we performed an experimental comparison against implementations run on a CPU. Our results with realistic cost models indicate that the move to GPUs can engender a speed-up of an order of magnitude.