Depth Map Refinement via Mutual Feature Weighting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current depth estimation methods in machine vision applications face challenges in achieving high accuracy and speed, particularly in real-time applications such as automatic vehicle braking, due to the complexity of stereo correspondence and the presence of noise and occlusions in image data.
Innovation Solution
The implementation of a stochastic approach combining iterative refinements and a sequence of processing blocks, including Local Matching, Cost Aggregation, Optimization, Mutual Structure, Dynamic Joint Weighted Median Filter, and Multi-Dimensional Convolution, to generate an optimized disparity map using the Adaptive Random Walk with Restart algorithm, addressing discontinuities and occlusions for improved accuracy and speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional stereo matching algorithms are used to construct disparity maps, then depth estimation can be achieved, but the accuracy is reduced due to noise and occlusions in image data
Solution Approach 1:
The patent segments the disparity map processing into multiple stages: initial disparity map construction, identification of discontinuity regions, and selective refinement in those regions. This segmentation allows the algorithm to focus computational effort on problematic areas while preserving accuracy in well-matched regions, effectively addressing noise and occlusion issues without processing the entire map uniformly.
Solution Approach 2:
The patent applies local quality by implementing region-specific processing where discontinuity regions identified in the initial disparity map receive specialized refinement treatment. The algorithm adapts its processing strategy based on local characteristics - applying more aggressive refinement in regions with occlusions or noise while maintaining efficiency in well-matched areas, thereby improving overall depth estimation accuracy.
2Measurement precision
If data intensive iterative computations are performed to achieve satisfactory accuracy, then depth mapping precision improves, but execution speed decreases which is unacceptable for real time applications
Solution Approach 1:
The patent performs preliminary action by first constructing an initial disparity map using efficient stereo matching algorithms before applying refinement. This initial map serves as a foundation that guides subsequent refinement steps, allowing the algorithm to focus computational resources only on regions that require improvement rather than processing the entire image data intensively, thus maintaining real-time execution capability while achieving satisfactory accuracy.
Solution Approach 2:
The patent implements partial action by applying iterative refinement only to identified discontinuity regions rather than the entire disparity map. This selective approach performs sufficient computation in critical areas to achieve the necessary accuracy improvement while avoiding excessive processing in already well-matched regions, thereby maintaining execution speed for real-time applications.
3Productivity
If the complexity of stereo correspondence processing is reduced for faster execution, then real time processing becomes feasible, but depth estimation accuracy deteriorates
Solution Approach 1:
The patent uses preliminary action by quickly generating an initial disparity map that provides a rough depth estimation. This initial map is then used to identify regions requiring refinement, allowing the system to achieve real-time processing throughput while directing more computational resources to specific areas where accuracy improvement is most needed, thus balancing speed and precision.
Solution Approach 2:
The patent introduces an intermediary step of identifying discontinuity regions based on the initial disparity map. This intermediary structure serves as a guide for subsequent refinement, enabling the system to maintain fast initial processing while systematically improving accuracy in problematic regions through targeted refinement operations.
4Productivity
If simple matching algorithms are used to reduce computational complexity, then processing speed increases, but the ability to handle occlusions and discontinuities is insufficient
Solution Approach 1:
The patent segments the processing into an initial fast matching phase that achieves computational speed, followed by a refinement phase that specifically targets discontinuity and occlusion regions. This segmentation allows the system to benefit from simple algorithms in the initial phase while applying more sophisticated handling only where needed, thus maintaining both speed and robustness.
Solution Approach 2:
The patent applies local quality by implementing different processing strategies in different regions: simple fast matching in well-behaved areas and sophisticated refinement with occlusion handling in regions identified as having discontinuities or occlusions. This localized approach ensures robustness where needed while maintaining overall computational efficiency.
Data Source
AI summary
A method for improving accuracy of depth map information derived from image data descriptive of a scene. In one embodiment Mutual Feature Map data are created based on initial disparity map data values and the image data descriptive of the scene. The Mutual Feature Map data are applied to create a series of weighting functions representing structural details that can be transferred to the first disparity values to restore degraded features or replace some of the first disparity values with values more representative of structural features present in the image data descriptive of the scene.