Point Cloud Data Hole Repair Aggregation Algorithm Based on Optimized Neural Network

Abstract

In order to solve the problem of cost cloud data and hole repair efficiency and accuracy, this article offers a study of integrated cloud network hole algorithm research based on optimal neural network. This paper first introduces a common point cloud hole-filling algorithm, provides a neural network-based point cloud blank filling algorithm, and introduces hotspot problems in a given algorithm, combined with the technology related to high-performance computing, the parallel optimization technology based on OpenMP and CUDA is adopted to accelerate the algorithm accordingly. Experiments have shown that the accuracy of pre- and postoptimization of CUDA-based algorithms varies. As the model and input point cloud data increase, the accuracy of the algorithm decreases slightly. However, when the data increases to 104 orders of magnitude, the rate of accuracy decline meets an inflection point and finally stabilizes to about 96.8% in the environment of 132457 data of monk model. The point cloud hole-filling algorithm based on the optimal neural network given in this article is highly predictable, can well repair incomplete point cloud holes, has good repair effect on point cloud holes, and can obtain high acceleration ratio, which can provide reference for practical engineering application.