Point Cloud Attribute Coding Using Geometry-Weighted Sums

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current point cloud compression techniques, such as Level of Details (LoD) and Region-Adaptive Hierarchical Transforms (RAHT), are inefficient for both dense and sparse point clouds, as they either fail to leverage local correlations in sparse clouds or are computationally demanding in dense clouds, limiting their effectiveness in real-time applications like VR and automotive vision.

Innovation Solution

The proposed method employs a top-down coding process using geometry-weighted sums and inter-depth prediction, applying transforms per node rather than per direction, which includes predictive coding and up-sampling to efficiently compress point cloud attributes, improving compression performance across both dense and sparse point cloud scenarios.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If Level of Details (LoD) coding is used for point cloud attributes, then the coding process is simple, but compression performance is poor for both dense and sparse point clouds

Engineering Contradiction:
Improvecoding simplicityVSAvoidcompression performance
Core Design Contradiction:
Ease of manufactureVSLoss of information

Solution Approach 1:

The patent segments the point cloud attribute coding into multiple depth levels (LOD layers), where each level is coded independently with appropriate prediction methods. This segmentation allows the system to handle different density regions differently, improving overall compression performance while maintaining coding simplicity through modular processing.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies different coding strategies to different regions of the point cloud based on local density characteristics. For dense regions, inter-attribute prediction is used, while for sparse regions, simpler prediction methods are applied. This local adaptation resolves the contradiction by optimizing compression performance for each region without significantly increasing overall coding complexity.

Inventive Principle:
Principle #3Local quality

2Loss of information

If Region-Adaptive Hierarchical Transforms (RAHT) are applied, then compression performance improves for sparse point clouds, but computational complexity increases significantly

Engineering Contradiction:
Improvecompression performanceVSAvoidcomputational complexity
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The patent divides the point cloud into multiple Level of Detail layers and applies transforms only to specific layers rather than processing the entire point cloud hierarchy. This segmentation reduces computational complexity by avoiding redundant transform operations while maintaining compression performance through selective application of RAHT to where it provides the most benefit.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies inter-attribute prediction selectively to only those attributes and regions where it provides significant compression benefit, rather than applying it universally. This partial action approach reduces computational complexity by avoiding unnecessary prediction calculations in regions or attributes where simple coding suffices, while maintaining compression performance in critical areas.

Inventive Principle:
Principle #16Partial or excessive action

3Loss of information

If existing compression methods are used for real-time applications, then processing speed is insufficient, but improving compression performance increases processing time

Engineering Contradiction:
Improvecompression performanceVSAvoidprocessing speed
Core Design Contradiction:
Loss of informationVSSpeed

Solution Approach 1:

The patent segments the point cloud into multiple LOD layers that can be processed and transmitted independently. This allows for progressive transmission where critical layers are processed first, enabling real-time applications to start with lower quality and improve over time. The segmentation enables parallel processing of different layers, improving overall processing speed while maintaining good compression performance.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary organization of point cloud data into LOD layers and pre-computes prediction models during the encoding phase. This preliminary action allows the decoder to rapidly reconstruct point clouds with minimal real-time computation, achieving both high compression performance and fast processing speed for real-time applications.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3942830B1Methods and devices for predictive point cloud attribute coding
Publication Date: 2024.09.04 BLACKBERRY LTD
  • EP3942830B1 patent drawingFigure 1
  • EP3942830B1 patent drawingFigure 2
  • EP3942830B1 patent drawingFigure 3

AI summary

Methods and devices for attribute coding for point clouds. In atop-down coding process a predicted geometry-weighted sum of attributes is determined for each occupied sub-volume of a volume based on a prediction operation that uses at least a geometry-weighted sum of attributes for the volume. The prediction operation involves up-sampling data from the preceding (parent) depth of the coding tree. The predicted geometry-weighted sums of attributes and corresponding original geometry-weighted sums of attributes are subtracted, in the attribute domain or in a transform domain to result in residual coefficients. The AC residual coefficients are coded in the bitstream. The transform used has a DC coefficient property, whereby the DC coefficient from transform of the geometry-weighted sums of attributes for the set of child sub-volumes is the geometry-weighted sum of attributes for the volume.