Three-dimensional point cloud processing method and device based on geometric unwrapping, and equipment

Abstract

The invention discloses a three-dimensional point cloud processing method and device based on geometric unwrapping, and equipment. The method comprises the steps of obtaining and processing point cloud data of an object; carrying out geometric unwrapping on the point cloud data so as to divide the point cloud data into points with large geometric changes and points with small geometric changes according to the geometric change degree; and inputting the point cloud data into a trained convolutional neural network for feature extraction and classification segmentation, the neural network model being used for extracting local features and global features of the point cloud data; and learning mutual supplementary information between the points with large geometric changes and the points with small geometric changes by using a geometric attention module. According to the invention, the accuracy of point cloud data classification tasks and the cross combination rate of segmentation tasks canbe improved.

Description

technical field [0001] The present invention relates to the technical field of three-dimensional data processing, and more specifically, relates to a method, device and equipment for processing three-dimensional point clouds based on geometric unwrapping. Background technique [0002] Convolutional neural network (CNN) is a kind of feed-forward neural network that includes convolutional calculations and has a deep structure. It consists of one or more convolutional layers (corresponding to traditional image processing filters), fully connected layers, and pooling layers. Composition, the convolutional neural network has the ability of representation learning. Its artificial neurons can respond to surrounding units within a part of the coverage area. The low-level network can extract low-level image features such as edges, lines, and corners. The high-level network can iteratively extract more complex features from low-level features. The parameters of the convolutional neur...

AI Technical Summary

Problems solved by technology

However, in this way, all points or local point clouds are entangled together for processing at the same time, and there is a large amount of redundant information, which makes the neural network unable to capture the most important geometric information.

Images