Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
DEEP LEARNING-BASED POINT CLOUD UPSAMPLING METHOD
Document Type and Number:
WIPO Patent Application WO/2021/232687
Kind Code:
A1
Abstract:
Disclosed in the present invention is a deep learning-based point cloud upsampling method, comprising: obtaining training data made up of a first quantity of sparse input points and a second quantity of dense input points; constructing a deep network model, used for respectively performing copying and curvature-based sampling operations on an initial feature vector extracted from the first quantity of sparse input points, obtaining a second quantity of intermediate feature vectors, performing a joining operation on every intermediate feature vector, inputting intermediate feature vectors having undergone said joining operations into a multi-layer perceptron, and determining sampling prediction points on the basis of sampling feature vectors output from the multilayer perceptron; training the deep network model until target functions determined by the sampling prediction points and the dense input points converge; testing the deep network model, and obtaining point cloud data of a test object after upsampling. The present method is able to convert a sparse point cloud into a curvature adaptive distribution-based dense point cloud, accurately representing contours of an object, and better facilitating 3D data expression, rendering, and visualization.

Inventors:
JIA KUI (CN)
LIN JIEHONG (CN)
CHEN KE (CN)
Application Number:
PCT/CN2020/125380
Publication Date:
November 25, 2021
Filing Date:
October 30, 2020
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
UNIV SOUTH CHINA TECH (CN)
International Classes:
G06T17/20
Foreign References:
CN111724478A2020-09-29
CN107862293A2018-03-30
CN110163799A2019-08-23
US20130151210A12013-06-13
US20180089536A12018-03-29
Other References:
YANG, BIN ET AL.: "Adaptive Up-Sampling Algorithm of Point Cloud Model", APPLICATION RESEARCH OF COMPUTERS, vol. 29, no. 6, 30 June 2012 (2012-06-30), pages 2354 - 2356, XP055870261, ISSN: 1001-3695
Attorney, Agent or Firm:
JIAQUAN IP LAW (CN)
Download PDF: