Dynamic point cloud position prediction method based on graph convolutional network

Abstract

The invention discloses a dynamic point cloud position prediction method based on a graph convolutional network, and the method comprises the steps: taking an original point cloud three-dimensional coordinate as an input, converting the original point cloud three-dimensional coordinate into a distance matrix, transmitting the distance matrix into a graph neural network, predicting the distance matrix after denaturation, and calculating a corresponding three-dimensional coordinate, thereby completing the prediction of a dynamic point cloud position. Meanwhile, the graph neural network is composed of two sets of alternate edge convolution layers and point convolution layers and a distance matrix layer, edge features and point features of the dynamic point cloud are equally considered, edge convolution and point convolution are alternately performed in the neural network, and therefore the accuracy of dynamic point cloud position prediction can be effectively improved, and prediction errors are reduced. In conclusion, the method has the characteristic that the prediction precision can be effectively improved.

Description

technical field [0001] The invention relates to a dynamic point cloud position prediction method, in particular to a dynamic point cloud position prediction method based on graph convolution network. Background technique [0002] Point cloud deformation prediction has important practical significance in the fields of gestures, expressions, attitudes, molecular dynamics, and protein folding. The point cloud can be regarded as a graph network, and the operation on the point cloud needs to satisfy translation invariance, rotation invariance and displacement invariance. At present, for the prediction of dynamic point cloud position, graph neural network is generally used for prediction, but the existing graph neural network only considers the characteristics of points and ignores the characteristics of edges; but in many problems, the neighbor point of a point to the point The distance of has an important influence on the characteristics of the point, which will lead to low acc...

AI Technical Summary

Problems solved by technology

At present, for the prediction of dynamic point cloud position, graph neural network is generally used for prediction, but the existing graph neural network only considers the characteristics of points and ignores the characteristics of edges; but in many problems, the neighbor point of a point to the point The distance has an important influence on the characteristics of the point, which will lead to low accuracy when using the conventional graph neural network to predict the position of the dynamic point cloud

Therefore, the existing technology has the problem of low prediction accuracy

Images