Adaptive Support Regions for Panoramic Stereo Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional stereo matching techniques using fixed support regions fail when applied to panoramic cameras due to image distortion, leading to erroneous disparity maps and depth information.
Innovation Solution
The method involves adjusting corresponding fixed support regions into adaptive support regions based on a predetermined mapping relationship, using equirectangular projection and circumscribed plane projection to account for distortion in panoramic camera images, allowing for accurate matching cost computation and disparity information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If fixed support regions are used for stereo matching in panoramic camera images, then the matching process is simple and fast, but the matching accuracy deteriorates due to image distortion
Solution Approach 1:
The patent applies dynamics by transforming the fixed support region concept into adaptive support regions. The support regions are dynamically adjusted based on the distortion characteristics of panoramic camera images, allowing the matching process to adapt to varying distortion levels across different image regions while maintaining computational efficiency.
Solution Approach 2:
The patent implements local quality by applying different support region strategies to different parts of the image. Instead of using uniform fixed support regions throughout, the method adjusts support region characteristics locally according to the distortion patterns specific to each region of the panoramic image, thereby improving overall matching accuracy.
2Measurement precision
If perspective transformation is applied to remove mismatches in fisheye stereo camera, then the feature points align better, but the processing complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-establishing the mapping relationship between pixels in panoramic camera images before performing stereo matching. This preprocessing step creates a foundation that simplifies subsequent matching operations, as the distortion compensation is already incorporated into the pixel correspondence relationships.
Solution Approach 2:
The patent introduces an intermediary mapping relationship that acts as a mediator between the distorted panoramic images. This mapping relationship serves as an intermediate representation that preserves geometric accuracy while simplifying the matching process, avoiding the need for complex perspective transformations.
Data Source
Figure 1~2B
Figure 3~4
Figure 5~6
AI summary
Matching cost computation method and device are disclosed. The method comprises an obtainment step of obtaining a reference projection image and a target projection image, the reference projection image being an equirectangular projection image corresponding to a reference stereo image, the target projection image being an equirectangular projection image corresponding to a target stereo image; a determination step of determining, based on a predetermined mapping relationship, a first pixel and a first neighboring region of the first pixel on the reference projection image; a selection step of selecting, based on the predetermined mapping relationship and the first pixel on the reference projection image, a second pixel and a second neighboring region of the second pixel on the target projection image; and a computation step of computing, based on the first pixel and the first neighboring region as well as the second pixel and the second neighboring region, a matching cost between the first pixel and the second pixel.