Stereo Matching Noise Weighting for Depth Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing stereo matching processes, particularly those using the census transform method, face challenges in accurately calculating depth information due to noise, especially in flat areas, and struggle to adapt to varying environmental conditions and noise characteristics.
Innovation Solution
An image processing device and method that performs a census transform process on two images, calculates comparison result information, and sets weight information to adjust the contribution of neighboring pixels based on noise levels, distinguishing between pixels heavily and lightly affected by noise to improve robustness and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a census transform process is used for stereo matching, then depth information can be calculated, but noise in flat areas reduces measurement precision
Solution Approach 1:
The patent applies local quality by differentiating the treatment of neighboring pixels based on their individual noise characteristics. Each neighboring pixel is evaluated separately to determine its noise level, and weights are assigned accordingly - pixels with low noise receive higher weights while pixels with high noise receive lower weights. This localized differentiation allows the census transform to maintain precision in reliable areas while suppressing noise in problematic flat regions.
Solution Approach 2:
The patent changes the parameter of pixel contribution by introducing variable weights that dynamically adjust the influence of each neighboring pixel on the disparity calculation. Instead of treating all neighboring pixels uniformly, the system modifies their contribution parameters based on measured noise levels, thereby adapting the census transform to local image conditions and improving depth measurement accuracy.
2Measurement precision
If uniform weight is assigned to all neighboring pixels, then calculation is simple, but accuracy decreases when noise levels vary across pixels
Solution Approach 1:
The patent implements preliminary action by pre-evaluating the noise level of each neighboring pixel before performing the main disparity calculation. The system calculates noise metrics for all neighboring pixels in advance, uses these evaluations to determine appropriate weights, and then applies these pre-determined weights to the census transform. This preliminary assessment allows the system to adapt to varying noise conditions without adding significant complexity to the core calculation process.
3Reliability
If all neighboring pixels are used in census transform, then processing is efficient, but noise propagation reduces reliability
Solution Approach 1:
The patent applies local quality by evaluating each neighboring pixel's reliability individually and assigning different weights based on local noise characteristics. Pixels in reliable regions contribute more to the disparity calculation, while pixels in noisy flat areas contribute less. This localized quality assessment maintains processing efficiency by still using all neighboring pixels, but improves reliability by differentiating their contributions based on local conditions.
Solution Approach 2:
The patent converts the harmful effect of noise into a beneficial filtering mechanism by using noise level measurement to determine pixel weights. The presence of noise in flat areas, which traditionally degrades stereo matching, is transformed into useful information that guides the weight assignment process. By measuring noise and systematically reducing the influence of noisy pixels, the system turns a previously harmful factor into a criterion for improving calculation reliability.
Data Source
AI summary
An image processing device includes a processor including hardware, the processor being configured to calculate comparison result information through a comparison process performed on the results of a census transform process performed on a first image and the results of the census transform process performed on a second image, set weight information, calculate a cost using the comparison result information and the weight information, calculate the amount of disparity using the cost, set a first weight having as low degree of contribution to the cost on a neighboring pixel which is determined to be affected by noise to a large extent among a plurality of neighboring pixels, and set a second weight having a high degree of contribution to the cost on a neighboring pixel which is determined to be affected by noise to a small extent among the plurality of neighboring pixels.