Disparity Map Estimation Using Varying Aggregation Windows
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating disparity maps, such as local, global, and iterative approaches, face inefficiencies in computational complexity and sensitivity to occluded and textureless regions, particularly in improving the efficiency of iterative techniques like coarse-to-fine methods.
Innovation Solution
An image processing device employs varying size aggregation windows during initial iterations and joint bilateral up-sampling during subsequent iterations to refine disparity maps, reducing computational complexity and memory requirements while minimizing incorrect matches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If global approaches are used to generate disparity map, then sensitivity to local regions such as occluded and textureless regions is reduced, but computational cost is high
Solution Approach 1:
The patent segments the disparity map generation process into multiple iterative stages (coarse-to-fine approach), where each stage operates at different resolution levels. This segmentation allows the system to use simpler local methods at coarse levels and more accurate global methods at fine levels, reducing overall computational cost while maintaining reliability in occluded and textureless regions.
Solution Approach 2:
The patent applies preliminary action by performing disparity estimation at coarser resolution levels first, establishing an initial disparity map that serves as a foundation for subsequent refinement. This preliminary estimation reduces the search space and computational burden for later fine-level processing, while still capturing global constraints that improve reliability in difficult regions.
2Productivity
If iterative approaches are used to improve efficiency, then computational efficiency is improved, but complexity of the process increases
Solution Approach 1:
The patent implements dynamics by adaptively adjusting the aggregation window size during iterative processing. The window size varies dynamically based on the iteration stage and local image characteristics, allowing the system to optimize computational efficiency at each stage while managing process complexity through systematic control rules.
Solution Approach 2:
The patent changes parameters systematically across iterative stages, including aggregation window size, pyramid level resolution, and processing thresholds. These parameter changes are orchestrated to improve computational efficiency at each stage while the overall process complexity is managed through the structured coarse-to-fine framework that guides parameter transitions.
3Manufacturing precision
If high-resolution images are processed in real-time, then image quality is maintained, but computational complexity and memory requirements increase
Solution Approach 1:
The patent transitions to another dimension by processing images at multiple resolution levels (pyramid levels) rather than directly at full resolution. This dimensional approach allows real-time processing of high-resolution images by working through intermediate lower-resolution representations, maintaining final image quality while reducing computational complexity and memory requirements at each processing stage.
Solution Approach 2:
The patent applies preliminary action by creating and processing a pyramid of downsampled image representations before final disparity map generation. This preliminary processing at lower resolutions reduces the computational burden and memory requirements, while the results are subsequently upsampled and refined to maintain high-resolution output quality.
Data Source
AI summary
An image processing device may include a disparity map estimator, which may perform an iterative technique to improve the efficiency of generating disparity map. The disparity map estimator performs disparity map refinement using varying size aggregation windows during the first iterations. The disparity map estimator uses larger size aggregation windows during the initial iterations to perform disparity map refinement. Further, the disparity map estimator may use joint bilateral up-sampling during the second (or last) iterations. Using a combination of varying size aggregation windows during the initial iterations and a joint bilateral up-sampling during the second iterations may reduce the computational complexity and memory requirements. In one embodiment, the combination of using varying size aggregation windows and joint bilateral up-sampling to refine the disparity map may reduce the high resolution of the images to a level acceptable for real-time processing and may further reduce the number of incorrect matches.