Adaptive Pixel Mask Selection for Stereo Depth Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing stereoscopic imaging systems face challenges in selecting the right pixel mask for matching processes, leading to incorrect depth calculations due to occlusion and background-foreground separation issues.
Innovation Solution
Adapting geometrical parameters of pixel masks to suit the optical system, projected pattern, and image sensor characteristics, and selecting masks with minimal signal variance to enhance matching accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a fixed mask is used for all pixels, then the device complexity is reduced, but the measurement precision deteriorates in occlusion and contour areas
Solution Approach 1:
The patent implements dynamic mask selection by evaluating multiple candidate masks (different sizes and shapes) for each pixel based on local image characteristics such as gradient magnitude and variance. This allows the system to adapt the mask configuration to local features like occlusion boundaries and contours, improving depth calculation accuracy without requiring a completely complex predetermined system.
Solution Approach 2:
The system changes mask parameters (size, shape, orientation) based on local image properties. Specifically, it selects from multiple mask types including circular, square, and rectangular masks with varying dimensions, choosing the optimal parameters for each pixel's local region to enhance measurement precision while maintaining manageable system complexity.
2Loss of information
If a large matching window is used, then more information is available for matching, but wrong matching results occur in occlusion areas
Solution Approach 1:
The patent segments the matching process by dividing the image into different regions (occlusion areas, non-occlusion areas, contour areas) and applying appropriate mask selections for each segment. This segmentation allows the system to use larger windows where appropriate while restricting window size in occlusion areas, thereby maintaining information availability where reliable while preventing wrong matches in problematic regions.
Solution Approach 2:
The system applies different mask qualities and sizes to different local regions based on their characteristics. In occlusion-free areas with sufficient texture, larger masks are used to maximize information availability. In occlusion and contour areas, smaller or differently shaped masks are selected to maintain reliability, thus achieving local optimization of both information usage and matching accuracy.
3Reliability
If standard deviation thresholding is applied, then matching reliability is improved, but processing time increases
Solution Approach 1:
The patent applies standard deviation thresholding selectively rather than uniformly across all pixels. It focuses the computationally intensive thresholding operation on critical regions such as occlusion boundaries and contours where reliability is most needed, while using simpler or pre-determined mask selections in homogeneous regions, thus improving overall depth map quality without proportionally increasing processing time.
Data Source
AI summary
A computational platform and a method for use in a process of matching pairs of pixels, wherein each of the members of a pixel's pair belong to another image captured by a different image capturing sensor, and wherein the computational platform comprises at least one processor configured to carry out a process of matching pairs of pixels based on selecting a pixel mask to be used by selecting neighboring pixels of a given pixel from among all of its neighboring pixels, will be used for the matching process of said given pixel with the other member of its pixel's pair.