Point Cloud Residual Prediction for Lower G-PCC Coding Bits
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The coding efficiency of point clouds in Geometry Point Cloud Compression (G-PCC) is affected by the overhead associated with coding residual data, particularly when predicting radius, azimuth, and laser index, as existing methods do not effectively utilize residual prediction to minimize bit usage.
Innovation Solution
Implementing residual prediction in G-PCC encoding and decoding processes to predict residual radius, azimuth, and laser index based on already-coded points, allowing for more efficient coding by using fewer bits to represent the remaining residual values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If residual prediction is applied to predict residual radius, azimuth, and laser index, then coding efficiency is improved by reducing the number of bits required, but the device complexity increases due to additional prediction steps and processing requirements
Solution Approach 1:
The patent applies preliminary action by predicting residual values (residual radius, azimuth, and laser index) before the actual coding process. The G-PCC coder determines predicted residual values based on already-coded points and uses these predictions to reduce the magnitude of residuals that need to be coded, thereby reducing the number of bits required for coding while maintaining a structured approach to manage the additional processing steps
2Loss of information
If the predicted values are closer to the reconstructed values, then fewer bits are needed to code the residual, but the calculation complexity increases due to additional prediction operations
Solution Approach 1:
The patent applies parameter changes by operating in the (r, φ, i) domain (radius, azimuth, laser index) rather than traditional Cartesian coordinates. This parameter transformation enables more effective prediction of residual values, as the spherical coordinate system better represents the geometric structure of point clouds acquired by rotating laser scanners, leading to smaller residual magnitudes and reduced coding bits despite the additional coordinate transformation steps
Data Source
AI summary
A device for decoding a point cloud includes a memory configured to store data representing the point cloud and one or more processors implemented in circuitry and configured to: decode a residual value; determine whether residual prediction is to be applied to a current point of the point cloud; in response to determining that residual prediction is to be applied to the current point of the point cloud: determine a residual predictor and add the residual predictor to the residual value to determine a final residual value; determine a predictor used to predict a current point of the point cloud; and use the predictor and the final residual value to determine a position of the current point.