Point Cloud Geometry Coding for Spinning Sensor Residual Radii
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing point cloud codecs, such as G-PCC, struggle to efficiently compress sparse geometry data captured by a spinning sensor head due to the continuous nature of the radius, leading to inefficient entropy coding of residual radii, which constitutes a significant portion of the bitstream.
Innovation Solution
Entropy encoding/decoding the sign of the residual radius based on the sign of a previously encoded/decoded non-zero residual radius, using context-based entropy coding, and adaptively ordering points by sensor index and azimuthal angle to improve compression efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing point cloud codecs (G-PCC) are used to compress sparse geometry data captured by spinning sensor head, then the compression process can be performed, but the compression efficiency is low due to inefficient entropy coding of residual radii
Solution Approach 1:
The patent changes the parameter representation by separating the residual radius into sign and magnitude components. The sign is entropy-coded using context-based prediction from previously coded signs, while the magnitude is encoded separately. This parameter transformation enables more efficient compression by exploiting the statistical properties of residual radius signs in spinning sensor data.
Solution Approach 2:
The patent implements feedback-based entropy coding where the sign of the current residual radius is predicted based on the sign of previously coded residual radii. This context-based feedback mechanism adapts to the local characteristics of the point cloud data, improving compression efficiency by reducing the entropy of the encoded sign information.
2Measurement precision
If the radius is continuously represented in the point cloud data, then the geometric precision is maintained, but the bitstream size increases significantly
Solution Approach 1:
The patent segments the residual radius representation into two distinct parts: the sign (positive or negative) and the magnitude. This segmentation allows differential encoding strategies to be applied to each component, with the sign being entropy-coded based on contextual prediction and the magnitude being encoded separately, thereby reducing the overall bitstream size while preserving geometric precision.
3Productivity
If points are encoded in arbitrary order, then the encoding process is simple, but the compression efficiency is reduced due to lack of exploitable spatial correlations
Solution Approach 1:
The patent applies preliminary ordering of points based on sensor index and azimuthal angle before the entropy coding stage. This preliminary arrangement exploits the spatial correlations inherent in spinning sensor data, where points captured by the same sensor at similar azimuthal angles tend to have similar radial distances. This pre-ordering enables more effective context-based prediction in the subsequent entropy coding step.
Data Source
Figure 1~2
Figure 3
Figure 4~9
AI summary
There is provided methods and apparatus of encoding/decoding a point cloud representing a physical object, each point of the point cloud being associated with a residual radius between a radius responsive to a distance of the point from a referential and a predicted radius. The method entropy encodes/decodes the sign of the residual radius associated with a current point of the point cloud based on the sign of a previously entropy encoded/decoded non-zero residual radius associated with a previous point.