3D Point Cloud Quality Prediction With Multi-View Graph Convolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing point cloud quality assessment methods fail to effectively capture global structure information and suffer from redundant calculations, leading to limited prediction accuracy due to their reliance on projection-based methods that do not consider multi-view perception.
Innovation Solution
A 3D point cloud quality prediction method using a graph convolutional neural network (GCNN) that performs dual-path multi-view projection, preprocesses images, and employs a backbone module, multi-layer attention perception, multi-layer conversion, and quality prediction modules to construct graph structures, enhancing feature representation and reducing redundant calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If projection-based methods are used for point cloud quality assessment, then the assessment can be performed, but global structure information cannot be effectively captured and redundant calculations occur
Solution Approach 1:
The patent divides the point cloud data into multiple projection planes (front, back, left, right, top, bottom views) and processes each projection independently through separate convolutional neural network paths. This segmentation allows the system to capture global structural information from different perspectives while avoiding redundant calculations by processing each view separately and then fusing the features.
2Measurement precision
If six projection planes are used for quality assessment, then quality evaluation can be performed, but multi-view perception by human visual system is not considered
Solution Approach 1:
The patent transforms the traditional single-view or limited-view projection approach into a comprehensive six-view projection system, adding dimensional completeness by incorporating front, back, left, right, top, and bottom views. This multi-dimensional approach better simulates human visual system's multi-view perception capability and provides more comprehensive quality assessment.
3Measurement precision
If traditional point-based or projection-based methods are used, then quality assessment is achievable, but correlation between different projection images is not utilized
Solution Approach 1:
The patent merges features from six different projection planes by fusing their respective feature maps in the neural network. The feature fusion module combines correlated information from multiple views while maintaining the unique characteristics of each projection, achieving better quality assessment without excessive model complexity through efficient feature integration.
Data Source
AI summary
A 3D point cloud quality prediction method based on graph convolutional neural network is provided. The beneficial effect is that the present disclosure can effectively capture global structural information, reduce redundant calculations, and improve the accuracy of the predict quality score.