Large-scale point cloud semantic segmentation method based on superpoint graph

Abstract

The invention provides a large-scale point cloud semantic segmentation method based on a superpoint graph. The method mainly includes: geometric uniform region division, superpoint graph construction,superpoint embedding, semantic segmentation, training and testing, and comprises processes of dividing the point cloud into geometric shapes called superpoint; using the entire point cloud as an input by means of an unsupervised step; calculating the superpoint graph in the geometric divided regions; then selecting a fixed-size dimension in each superpoint; calculating a descriptor by embedding avector; and finally, because the graph of the superpoint is smaller than the graph created on the original point cloud, using a deep learning algorithm based on graph convolution to classify nodes with rich edge features, and embedding the superpoints according to the information transmitted by the super edges. The method solves the semantic segmentation on a large three-dimensional point cloud.The superpoint graph processes large-scale data on the basis of a deep learning framework, and improves the segmentation efficiency while retaining micro details.

Description

technical field [0001] The invention relates to the field of semantic segmentation, in particular to a large-scale point cloud semantic segmentation method based on a superpoint graph. Background technique [0002] Semantic segmentation is a machine that automatically segments and recognizes the content in an image. It can be said to be a basic technology for image understanding, and plays a pivotal role in automatic driving systems, drone applications, and wearable device applications. As we all know, an image is composed of many pixels, and "semantic segmentation", as the name suggests, is to segment pixels according to the different semantic meanings expressed in the image. Semantic segmentation is an important branch in the field of artificial intelligence, and it is an important aspect of image understanding in machine vision technology. An important part is that in the autonomous driving technology in recent years, after the on-board camera detects images of obstacles ...

AI Technical Summary

Problems solved by technology

Although there are many studies on point cloud semantic segmentation, the available data is small in size and has a fuzzy structure, which leads to the inefficiency of convolutional neural networks in processing images on irregular data, so semantic segmentation on large 3D point clouds challenges remain

Images