Multi-View Panoramic Stereo Model Generation With Fused Cost Volumes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional multi-view stereo vision methods consume excessive computing resources and suffer from low precision due to inefficient matching point searches, leading to poor accuracy in generated stereo vision models.
Innovation Solution
A model generation method and apparatus based on multi-view panoramic images that utilize image rectification, cost volume calculation and fusion, and phase difference estimation to improve precision while reducing resource consumption, employing techniques like folding and tiling dimensionality reduction and residual convolutional networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional multi-view stereo vision methods perform image sampling and matching point search, then a stereo vision model can be generated, but the consumption of computing resources increases significantly
Solution Approach 1:
The patent divides the image processing into multiple resolution levels (first-level reference image, lower-level reference images). By performing dimensionality reduction to create lower-resolution versions and processing these segmented levels separately, the computational burden is distributed and reduced, allowing model generation with lower computing resource consumption.
Solution Approach 2:
The patent introduces a resolution dimension by creating multiple levels of image abstraction (original resolution, lower resolution). This dimensional transformation allows the system to process images at different scales, reducing the computational complexity of matching point search while still generating accurate stereo vision models through multi-level fusion.
2Reliability
If traditional multi-view stereo vision methods perform matching point search, then a stereo vision model can be generated, but the precision of the generated model deteriorates due to low accuracy of matching points
Solution Approach 1:
The patent segments the matching process into multiple resolution levels. By performing matching at lower resolutions first and then refining at higher resolutions, the system achieves more accurate matching points. The multi-level approach allows coarse matching at low resolution followed by fine matching at high resolution, improving overall precision.
Solution Approach 2:
The patent performs preliminary dimensionality reduction to create lower-resolution versions of images before performing detailed matching. This preliminary action at reduced resolution provides a rough matching framework that guides subsequent high-resolution matching, improving the accuracy of final matching points while reducing overall computational complexity.
3Measurement precision
If image rectification and cost volume calculation are performed, then phase difference estimation accuracy is improved, but the complexity of the processing increases
Solution Approach 1:
The patent segments the complex processing into distinct modules: image rectification, feature extraction, cost volume calculation, and phase difference estimation. By dividing these processing steps and applying them at multiple resolution levels, the system manages complexity while maintaining high phase difference estimation accuracy through systematic multi-level processing.
Data Source
AI summary
The disclosure provides a model generation method based on a multi-view panoramic image, including: calculating an image rectification rotation matrix of source images and a reference image; extracting a reference image feature of the reference image and source image features of the source images; performing a fusion operation on rectified cost volumes of the plurality of source images corresponding to the reference image to obtain a final cost volume; calculating an estimated phase difference under a set resolution; obtaining a final phase difference of the reference image; and generating a depth map of the reference image, and constructing a corresponding stereo vision model.