Point Cloud Occupancy Coding with Neighbour-Weighted Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current point cloud compression methods are inefficient in encoding and decoding large datasets of three-dimensional objects, particularly in sparse populations, which hampers their adoption and deployment in applications like autonomous vehicles and virtual reality, due to challenges in coding geometry and spatial correlation.

Innovation Solution

The method involves context-adaptively coding occupancy data for child sub-volumes based on the occupancy status of neighboring sub-volumes, using weights that reflect the correlation between neighboring and child sub-volumes, and employing entropy encoding and decoding techniques to generate and reconstruct compressed bitstreams.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If traditional compression methods are used for point cloud data, then implementation is simpler, but compression efficiency is insufficient especially for sparse point clouds

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

Solution Approach 1:

The point cloud data is divided into multiple sub-volumes (e.g., octree structure) to enable localized processing. Each sub-volume is encoded independently using occupancy prediction, allowing the system to handle sparse regions more efficiently while maintaining overall compression performance.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary occupancy prediction for child sub-volumes based on parent sub-volume occupancy status before actual encoding. This preliminary action allows the encoder to prepare probability models and prediction structures in advance, improving compression efficiency without significantly increasing runtime complexity.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If more data is stored to maintain fidelity, then accuracy is improved, but storage requirements increase

Engineering Contradiction:
Improveoccupancy prediction accuracyVSAvoidstorage requirements
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent employs feedback mechanisms where occupancy prediction results from parent sub-volumes are used to inform the encoding of child sub-volumes. This feedback loop allows the system to maintain high prediction accuracy by continuously refining probability models based on previously encoded data, reducing the need to store redundant information.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The system dynamically adjusts encoding parameters such as probability thresholds and prediction models based on the occupancy characteristics of different sub-volumes. By changing parameters adaptively rather than using fixed settings, the patent achieves high accuracy while optimizing storage requirements for different regions of the point cloud.

Inventive Principle:
Principle #35Parameter changes

3Productivity

If compression is increased to reduce bandwidth, then transmission efficiency is improved, but decoding complexity increases

Engineering Contradiction:
Improvetransmission efficiencyVSAvoiddecoding complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent introduces intermediate prediction structures and probability models that act as mediators between the compressed bitstream and the reconstructed point cloud. These intermediaries enable efficient decoding by providing structured guidance for reconstructing occupancy data without requiring complex computational operations at the decoder side.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP3595180B1Methods and devices for neighbourhood-based occupancy prediction in point cloud compression
Publication Date: 2021.12.08 BLACKBERRY LTD
  • EP3595180B1 patent drawingFigure 1~2
  • EP3595180B1 patent drawingFigure 3~4
  • EP3595180B1 patent drawingFigure 5~6

AI summary

Methods and devices for encoding a point cloud, where occupancy data for child sub-volumes is context-adaptively encoded based on contexts selected, at least in part, using an occupancy score determined for each child sub-volume. The occupancy score for each child sub-volume of a parent sub-volume is determined based on the occupancy status of a plurality of neighbouring sub-volumes neighbouring the parent sub-volume and a respective weight assigned to each of the neighbouring sub-volumes vis-à-vis that child sub-volume, and wherein the respective weight reflects a strength of correlation between occupancy of that neighbouring sub-volume and occupancy of that child sub-volume.