Point Cloud Encoding Using Predictive Radius Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing point cloud compression technologies fail to combine encoding and decoding simplicity, low latency, and high compression performance effectively, particularly for sparse geometry data sensed by spinning sensors, which is crucial for applications like autonomous driving where real-time data processing is essential.
Innovation Solution
The method involves encoding point cloud geometry data by selecting a predicting radius from precedingly encoded radii associated with different sensor indices, allowing for improved prediction and reduced dynamic range of residual radii, thereby enhancing compression efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If traditional point cloud compression methods are used, then encoding and decoding processes are simple, but compression performance is poor and latency is high
Solution Approach 1:
The patent applies preliminary action by performing prediction using radii from previously encoded points before actual encoding occurs. The encoder maintains a list of previously encoded radii and uses them to predict current point radii, reducing the dynamic range of residuals that need to be encoded. This preliminary prediction step is crucial for achieving high compression performance while maintaining encoding simplicity.
Solution Approach 2:
The patent changes parameters by transforming the radius representation from absolute values to predicted residuals. By selecting prediction radii from previously encoded points and encoding only the difference (residual), the dynamic range of encoded values is significantly reduced. This parameter transformation enables more efficient entropy coding and achieves superior compression performance.
2Loss of time
If traditional point cloud compression methods are used, then encoding process is simple, but decoding latency is high
Solution Approach 1:
The patent applies preliminary action by pre-selecting prediction radii from previously decoded points during the decoding process. The decoder maintains the same list of previously decoded radii as the encoder and uses identical prediction logic, enabling real-time decoding without requiring complex post-processing. This preliminary prediction reduces the computational burden during decoding, lowering latency.
Solution Approach 2:
The patent implements feedback by using previously decoded radius information to inform current decoding decisions. The decoder maintains a list of previously decoded radii and uses this feedback to predict current point radii, reducing the dynamic range of residuals that need to be processed. This feedback mechanism enables efficient real-time decoding while maintaining high compression performance.
3Loss of energy
If prediction radius is selected only from same sensor index points, then encoding is simpler, but compression performance is reduced
Solution Approach 1:
The patent applies universality by enabling prediction across multiple sensor indices rather than restricting it to a single sensor. The prediction mechanism works universally with points from different sensors by selecting the best matching prediction radius from the list of previously decoded radii regardless of sensor origin. This multi-functional approach significantly improves compression performance while maintaining manageable complexity through systematic selection criteria.
Solution Approach 2:
The patent changes parameters by expanding the prediction search space to include radii from multiple sensor indices. Instead of limiting prediction to same-sensor points, the system considers radii from various sensors and selects the best match based on geometric proximity and temporal relevance. This parameter expansion improves prediction accuracy and compression performance while the systematic selection process keeps complexity controlled.
Data Source
AI summary
A method of encoding, into a bitstream, point cloud geometry data sensed by at least one sensor associated with a sensor index, includes: selecting, for a first occupied coarse point having a first sensor index and being associated with a first radius, a selected predicting radius from at least one second radius associated with at least one second occupied coarse point having a second sensor index different from the first sensor index, and from at least one third radius associated with at least one third occupied coarse point having a sensor index equal to the first sensor index; encoding, in the bitstream, data indicating if the selected predicting radius is equal to either a second radius or a third radius; and predictively encoding, in the bitstream, a residual radius between the first radius and the selected predicting radius.