Point Cloud Interpolation Decoding for Lower-Bitrate 3D Compression

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Point clouds require significant bandwidth for transmission due to their large bitrate, and existing video codecs are not effective in encoding and decoding 3D video content, leading to high computational costs and hardware expenses.

Innovation Solution

Convert point clouds into a 2D state by projecting them onto multiple frames, using existing video codecs to compress and reconstruct the 2D frames, and then transmit the bitstream for decoding and rendering.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If conventional point cloud compression methods are used, then implementation complexity is reduced, but compression efficiency is insufficient

Engineering Contradiction:
Improvecompression efficiencyVSAvoidimplementation complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The point cloud data is segmented into multiple patches, and each patch is processed independently through an autoencoder network. This segmentation enables parallel processing and improves compression efficiency without requiring complex global optimization algorithms.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies K-means clustering to pre-group points into patches before compression processing. This preliminary organization of data structure enables more efficient subsequent encoding and decoding operations, improving overall compression performance.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If more bits are allocated to each point, then point cloud quality is improved, but bitrate consumption increases

Engineering Contradiction:
Improvepoint cloud qualityVSAvoidbitrate
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent transforms point cloud data from spatial coordinates into latent space representations through autoencoder networks. This parameter transformation enables more efficient bit allocation, achieving high reconstruction quality at lower bitrates by capturing essential features in the latent space rather than storing raw coordinates.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The autoencoder network acts as an intermediary that compresses point cloud data into a compact latent representation. This intermediate form retains essential geometric information while requiring fewer bits for storage and transmission, enabling high quality reconstruction with reduced bitrate.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Productivity

If existing compression schemes are applied, then implementation is simpler, but they do not take advantage of 3D structure and are outperformed by 2D image codecs

Engineering Contradiction:
Improvecompression efficiencyVSAvoidscheme complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent processes 3D point cloud data by organizing points into 2D-like patches while preserving 3D spatial relationships through coordinate transformations and depth information. This approach enables the use of efficient 2D image coding techniques while maintaining 3D structural integrity, achieving superior compression compared to traditional 3D methods.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP3782122B1Point cloud compression using interpolation
Publication Date: 2026.04.29 SAMSUNG ELECTRONICS CO LTD
  • EP3782122B1 patent drawingFigure 1~2
  • EP3782122B1 patent drawingFigure 3
  • EP3782122B1 patent drawingFigure 4A

AI summary

A method for decoding point cloud includes decoding the compressed bitstream into a geometry frame representing geometry of a 3D point cloud and a texture frame representing texture of the 3D point cloud, the geometry frame and the texture frame including pixels representing points of the 3D point cloud from different layers of the 3D point cloud; deriving a set of missing geometry values from pixels in the geometry frame and a set of missing texture values from pixels in the texture frame; generating a first set of frames representing geometry from the different layers based on the geometry frame and the set of missing geometry values; generating a second set of frames representing the texture from the different layers based on the texture frame and the set of missing texture values; and generating the 3D point cloud using the first set of frames and the second set of frames.