Parallel Research On Point Cloud Hole Filling Algorithm Based On Machine Learning

As the science and technology of computer vision and image developing and progressing, the model of scattered point cloud data can be conveniently obtained through the scanning device. In the process of surface reverse reconstruction, due to the damage of the model or object occlusion, etc.,It will inevitably appear that point cloud data may be lost, which will directly affect the point cloud effect of subsequent processing. Accurate data for defect repair is the prerequisite of the reverse reconstruction. Therefore, the repair the defect of point cloud data is needed before modeling the scattered point cloud.Based on the study of domestic and foreign research, with scattered point cloud data model as the target object, this essay focuses on the process of scattered point cloud hole repairing and the algorithm of radial basis function (RBF) based on machine learning. It can be directly fitted in terms of the implicit surface equation of the scattered point cloud and thus make an effective repair to the defect data appeared in the process of measurement. Due to a large number of point cloud data will increase the amount of numeration, in order to accelerate the algorithm, it should be considered that the current parallel technology (such as multi-core acceleration (OpenMP), GPU (CUDA), and CPU/ GPU hybrid parallel technology) will enhance the algorithm so as to shorten the time running as well as increase the efficiency. Through the efficiency analysis of the experimental results, it is proved that the algorithm can be parallel, and a certain algorithm is obtained. The main works of this paper are as follows:1. The basic process of conventional hole filling and radial basis function interpolation process is studied, the algorithm based on multi scale radial basis function is realized.2. The improvement of serial algorithm in multi-core CPU, analysis and design of the OpenMP-based multi-core parallel algorithm, and in different environments for the experimental results analysis, the algorithm has achieved good acceleration efficiency.3. The design and implementation of the CUDA-based parallel algorithm, and in different environments for the analysis of the experimental results, the algorithm has achieved a good acceleration.4. The design and implementation of a hybrid parallel algorithm based on OpenMP and CUDA, and in different experimental environment for the test results analysis, the results show that the proposed algorithm to accelerate the ratio has been further improved.