3D Point Cloud Compression via Multi-Scale Dictionary Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 3D point cloud compression technologies face challenges in efficiently compressing massive data, particularly in real-time applications like autonomous driving, due to high computational complexity and inadequate characterization of complex structural characteristics, leading to degraded compression quality.
Innovation Solution
A 3D point cloud compression system based on multi-scale structured dictionary learning, which partitions point cloud data into voxels and blocks, learns a multi-scale dictionary through hierarchical sparse coding, and performs adaptive encoding and decoding to improve compression efficiency and reconstruction quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If graph Fourier transform is used to compress point cloud data, then compression capability is improved, but computational complexity increases significantly
Solution Approach 1:
The patent segments the point cloud data into multiple blocks and processes each block independently using localized graph Fourier transform. This divides the large-scale eigenvalue decomposition problem into smaller sub-problems, reducing computational complexity while maintaining compression capability.
Solution Approach 2:
The patent changes the parameter of transform basis from globally calculated Laplacian eigenvectors to locally calculated basis functions within each block. This parameter change reduces the scale of matrix operations from the entire point cloud to individual blocks, significantly lowering computational complexity.
2Productivity
If fixed analytic transforms are used, then encoding efficiency is improved, but adaptability to attribute signals deteriorates
Solution Approach 1:
The patent introduces dynamic adaptability by learning transform basis matrices that are specific to different attribute signals (color, texture, reflectivity) rather than using a single fixed transform. Each attribute type can have its own optimized basis, improving both encoding efficiency and adaptability.
Solution Approach 2:
The patent applies different transform basis matrices to different attribute signals based on their specific characteristics. Each attribute type receives a customized transform that matches its statistical properties, achieving local optimization for each attribute while maintaining overall encoding efficiency.
3Productivity
If 3D-2D projection is used to compress point cloud, then compression is achieved, but projection distortion is introduced
Solution Approach 1:
Instead of projecting 3D point cloud data onto a 2D plane, the patent performs compression directly in the 3D space using graph Fourier transform on the volumetric data. This maintains the three-dimensional structure and spatial relationships, avoiding projection distortion while achieving compression.
4Device complexity
If spatial transform coding depending only on geometric information is used, then encoding is simplified, but compression quality of attribute signal deteriorates
Solution Approach 1:
The patent changes the transform parameters by incorporating statistical characteristics of attribute signals into the basis matrix calculation. Instead of using only geometric information, the transform basis is adapted to match the statistical properties of each attribute type, significantly improving compression quality.
Solution Approach 2:
The patent uses feedback from attribute signal statistics to adjust and optimize the transform basis matrices. By analyzing the statistical characteristics of attribute signals and using this information to refine the transform parameters, the system achieves better compression quality while managing encoding complexity.
Data Source
AI summary
In a 3D point cloud compression system based on multi-scale structured dictionary learning, a point cloud data partition module outputs a voxel set and a set of blocks of voxels of different scales. A geometric information encoding module outputs an encoded geometric information bit stream. A geometric information decoding module outputs decoded geometric information. An attribute signal encoding module outputs a sparse coding coefficient matrix and a learned multi-scale structured dictionary. An attribute signal compression module outputs a compressed attribute signal bit stream. An attribute signal decoding module outputs decoded attribute signals. A 3D point cloud reconstruction module completes reconstruction. The system is applicable to lossless geometric and lossy attribute compression of point cloud signals. Based on the natural hierarchical partitioning structure of point cloud signals, the system gradually improves the reconstruction quality of high-frequency details in the signals from coarse scale to fine scale, and achieves significant gains.