3D point cloud encoding and decoding method, compression method and device based on graph dictionary learning

Patent metadata

Field	Value
Publication number	US-12046009-B2
Application number	US-202418590993-A
Country	US
Kind code	B2
Filing date	Feb 29, 2024
Priority date	Mar 22, 2022
Publication date	Jul 23, 2024
Grant date	Jul 23, 2024

Abstract

A graph dictionary learning method for a 3D point cloud comprises: obtaining N point clouds to form training dataset; performing voxelization process on the point cloud data to obtain voxelized point cloud data of the training dataset; performing voxel block division on the point cloud data of the training dataset, selecting a plurality of voxel blocks as the training dataset, and constructing a graph dictionary learning model according to the training dataset; and performing iterative optimization on the graph dictionary learning objective function to obtain a graph dictionary for encoding and decoding a 3D point cloud signal. The present disclosure effectively uses the spatial correlation between point cloud signals to near-optimally remove the redundancy among point cloud signals.

First claim

What is claimed is:

1. A graph dictionary learning method for a 3D point cloud, comprising:
obtaining a training dataset of N point cloud data;
performing voxelization process on the N point cloud data of the training dataset to obtain voxelized point cloud data of the training dataset;
performing voxel block division on the voxelized point cloud data of the training dataset, selecting a plurality of voxel blocks as a training dataset for the graph dictionary learning, constructing a graph dictionary learning model according to the training dataset, and establishing a graph dictionary learning objective function; and
performing iterative optimization on the graph dictionary learning objective function to obtain a graph dictionary for encoding and decoding a 3D point cloud signal;
wherein the performing voxel block division on the voxelized point cloud data of the training dataset and selecting a plurality of voxel blocks as a training dataset for the graph dictionary learning comprises:
uniformly dividing a boundary cube where all the voxelized point cloud data of the training dataset is located into m×m×m voxel blocks, m being a pre-set side length;
sorting all the voxel blocks of each point cloud data of the training dataset in descending order according to the number of contained voxels;
selecting top r voxel blocks containing the most number of voxels from each point cloud data of the training dataset, and calculating an average value of the attribute signals of the voxels contained in each voxel block in the top r voxel blocks as a direct current attribute signal of the voxel block, r being a pre-set positive integer; and
subtracting the direct current attribute signal from each voxel to obtain a residual attribute signal as a training signal, forming a training dataset for graph dictionary learning.

2. A graph dictionary learning method for a 3D point cloud, comprising:
obtaining a training dataset of N point cloud data;
performing voxelization process on the N point cloud data of the training dataset to obtain voxelized point cloud data of the training dataset;
performing voxel block division on the voxelized point cloud data of the training dataset, selecting a plurality of voxel blocks as a training dataset for the graph dictionary learning, constructing a graph dictionary learning model according to the training dataset, and establishing a graph dictionary learning objective function; and
performing iterative optimization on the graph dictionary learning objective function to obtain a graph dictionary for encoding and decoding a 3D point cloud signal;
wherein the constructing a graph dictionary learning model according to the training dataset, and establishing a graph dictionary learning objective function comprises:
constructing a topology-connected graph structure =(V, ε, W) by regarding a signal in the training dataset as a graph signal, wherein V represents a set with m 3 nodes; ε represents a set of edges connecting m 3 nodes; W∈ m 3 ×m 3 represents a weight matrix of the edges;
constructing a graph dictionary learning model by using the eigen-basis function of p-Laplacian operator of a topology-connected graph structure to establish a graph dictionary learning objective function:

min

D
,
A

1
n

⁢

∑

i
=
1

n

1
2

⁢



M
i

(

x
i

-

D
⁢

α
i

)



2
2

+

β
⁢

∑

s
=
1

s

J
⁡
(

D
s

)

+

γ
⁢



D
T

⁢
D

-
I



F
2

+

λ
n

⁢

∑

i
=
1

n



α
i



1

where x i ∈ m 3 represents an i th training voxel block; M i ∈ m i ×m 3 represents a mask matrix with each element having a value of 0 or 1 used for extracting a corresponding m i voxels from x i ; D∈ m 3 ×m 3 represents an overcomplete graph dictionary composed of S complete sub-dictionaries, namely, D=[D 1 , . . . , D S ];

J
⁡
(

D
s

)

=

∑

k

⁢

∑

i
⁢
j

Assignees

• Univ Shanghai Jiaotong

Inventors

• Dai Wenrui
• Li Xin
• Li Shaohui
• Li Chenglin
• Zou Junni
• Xiong Hongkai

Classifications

• Y02T10/40

  Engine management systems · CPC title

• G06T9/40 Primary

  Tree coding, e.g. quadtree, octree · CPC title

• G06T9/001 Primary

  Model-based coding, e.g. wire frame · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T2207/10028

  Range image; Depth image; 3D point clouds · CPC title