Confidence-Guided Stereo Disparity for Difficult Image Regions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional stereo algorithms struggle to accurately derive disparity information in scenarios with repetitive patterns, low illumination, contrast, and large depth discontinuities, while AI-based methods face challenges with semantically confusing scenes, requiring excessive computational resources.
Innovation Solution
A hybrid approach combining conventional and AI-driven disparity algorithms, where the former provides initial disparity information with a confidence score, and the latter supplements it in low-confidence regions, ensuring accurate and efficient disparity retrieval.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If conventional stereo-matching algorithms are used to derive disparity information, then the method is computationally efficient and easy to implement, but accuracy deteriorates in regions with repetitive patterns, low illumination, contrast, or large depth discontinuities
Solution Approach 1:
The patent merges conventional stereo-matching algorithms with AI-based algorithms into a hybrid system. The conventional algorithm provides initial disparity information and confidence scores, while the AI algorithm processes regions with low confidence scores to improve accuracy. This combination allows the system to maintain computational efficiency of conventional methods while achieving the accuracy improvements of AI methods in challenging regions.
2Measurement precision
If AI-based algorithms are used to derive disparity information, then accuracy improves in challenging regions, but computational resource requirements increase excessively
Solution Approach 1:
The patent applies local quality by using the AI algorithm only in specific regions where it is most needed - namely, regions with low confidence scores identified by the conventional algorithm. Instead of applying the computationally expensive AI algorithm to the entire image, the system selectively applies it to problematic regions such as those with repetitive patterns, low illumination, or large depth discontinuities, thereby reducing overall computational resource consumption while maintaining accuracy where it matters most.
Solution Approach 2:
The patent segments the image processing task into two parts: first, the conventional algorithm processes the entire image to generate initial disparity information and confidence scores; second, the AI algorithm processes only the segmented regions with low confidence scores. This segmentation approach allows the system to leverage the strengths of both algorithms while minimizing the computational burden of the AI component.
3Use of energy by moving object
If only conventional algorithms are used, then computational resources are conserved, but reliability deteriorates in scenarios with repetitive patterns, low illumination, and large depth discontinuities
Solution Approach 1:
The patent implements feedback by using the confidence score function generated by the conventional algorithm to guide the AI algorithm. The confidence scores provide feedback about which regions are unreliable, allowing the system to target AI processing to those specific regions. This feedback mechanism ensures that computational resources are allocated to improve reliability where it is most needed, rather than uniformly across the entire image.
4Measurement precision
If AI algorithms are applied to entire stereo image pairs, then comprehensive disparity information is obtained, but processing time increases excessively for fast applications
Solution Approach 1:
The patent applies partial action by using the AI algorithm only partially - specifically, only for regions with low confidence scores rather than the entire image. This partial application of the AI algorithm provides sufficient disparity information for challenging regions while avoiding the excessive processing time that would result from applying AI to the complete image, making the system suitable for fast applications such as autonomous navigation.
Data Source
AI summary
A method and a system for retrieving disparity information from a stereo image pair including a first and second images obtained by first and second cameras, respectively. The first and second images are provided to a conventional disparity algorithm to an artificial intelligence (AI) driven disparity algorithm. Using the conventional disparity algorithm, the first disparity information is derived from the stereo image pair, and a confidence score function is associated with the derived first disparity information. Using the AI-driven disparity algorithm, second disparity information is derived from at least part of the stereo image pair. The first and second disparity information are fused by giving priority to the first disparity information in regions of the stereo image pair with a high confidence score and giving priority to the second disparity information in regions of the stereo image pair with a low confidence score, thereby obtaining resultant disparity information.