A Research on CUBE Filtering Algorithm Based on GPU Parallel Acceleration

This paper focuses on execution efficiency for Combined Uncertainty and Bathymetry Estimator (CUBE) algorithm. By analyzing the key steps of the algorithm, this paper shows how to improve algorithm efficiency by using Octree index and Graphics Processing Unit (GPU) parallel computing. The former is used to establish rules grids and neighborhood searches, while the latter is used to calculate the filtering calculation task in parallel. The combination of the two improves the overall execution efficiency of the algorithm. This paper designs experiment to test efficiency of the algorithm when processing the Multi-Beam Echo Sounder (MBES) data, and compare with the CUBE filter module of CARIS HIPS software. The experimental results show that the algorithm used in this paper can increase the filtering speed by about 13.8 times when dealing with ten million levels of data, and the efficiency of this algorithm in processing millions of MBES data is significantly better than the efficiency of CARIS HIPS software; The algorithm has better denoising effect than CARIS HIPS software in some cases.