Parallel Research On Surface Reconstruction Algorithm Based On Multi Hidden Layer Back Propagation Neural Network

With the gradual development and maturity of 3D surface reconstruction technology, people have been optimizing and improving the technology in order to reconstruct the high precision object surface,and applied it to many fields. Among them, the application of BP neural network to object surface reconstruction is also an important means to improve surface reconstruction technology, and the training accuracy of BP network will directly affect the accuracy of surface reconstruction. Therefore,the BP network can be improved to improve the training accuracy of the network, so as to better reconstruct the surface of the object.This paper improves the training accuracy of the network by increasing the number of hidden layers in the BP network. However, increasing the number of hidden layers in the network will dramatically increase the time taken by the algorithm. In addition, this paper mainly for surface reconstruction of point cloud data, the number of point cloud data is very large, and the method for all data points will operate, thus reducing the overall execution time of the algorithm is an urgent problem.Parallel computing is an important method to solve this problem. It can effectively reduce the execution time of the algorithm under the condition that the original precision of the serial algorithm is guaranteed. This paper mainly uses multi-core parallel technology and GPU based parallel technology to reduce the overall time consuming and improve the efficiency of the algorithm.This algorithm can be summarized as follows: firstly to obtain point cloud information from a text file, using the octree to store these point cloud data, input matrix and output matrix of the network, and then through the multi hidden layer BP neural network to the point cloud data for training, so as to obtain the implicit function surface, the surface is the approximation of implicit function the real object surface, the implicit function in the network storage weight, according to the network weights in the implicit information using implicit function surface visualization method shows the object surface.Then parallel acceleration is performed for the more time-consuming part of the serial algorithm.The experimental results show that the proposed surface reconstruction algorithm based on multi hidden layer BP neural network can effectively improve the reconstruction accuracy of the object surface. Parallel computing technology can also shorten the execution time of the algorithm and improve the efficiency of the algorithm.