Stereo Image Matching with 3D-Guided Search Region Masking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for obtaining three-dimensional shape information from multiple images using stereo matching can result in inaccurate parallax information due to objects being shielded or out of view, leading to incorrect distance measurements, especially with irregularly shaped objects or multiple objects in the scene.
Innovation Solution
An image processing apparatus that uses hardware processors and memory to obtain captured images, calculate image capturing parameters, and set a search target region based on three-dimensional shape data, limiting the search range to improve parallax estimation by identifying valid corresponding points.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the search range is set to the entire second captured image to ensure comprehensive search, then the probability of finding the corresponding point increases, but the computational load and time required for stereo matching increase significantly
Solution Approach 1:
The patent applies local quality by creating a search target region mask that differentiates between searchable and non-searchable areas in the second captured image. Instead of uniformly searching the entire image, the system selectively applies search restrictions based on local image characteristics such as occlusion regions and out-of-viewing angle areas, thereby reducing computational load while maintaining accuracy in relevant regions
Solution Approach 2:
The patent performs preliminary action by generating the search target region mask before executing the stereo matching process. This mask is created by identifying occlusion regions and out-of-viewing angle areas in advance, allowing the system to pre-def ine the search boundaries and avoid unnecessary computations during the actual matching phase
2Reliability
If the search is performed in occluded regions or out-of-viewing angle areas, then more corresponding points may be found, but incorrect parallax information is obtained due to shielded or cut-off corresponding points
Solution Approach 1:
The patent converts the harmful effect of occlusion and out-of-viewing angle problems into a beneficial filtering mechanism. By identifying these problematic regions and excluding them from the search target, the system prevents the acquisition of incorrect parallax information. The harm of limited search coverage is transformed into the benefit of higher measurement precision by avoiding erroneous matches
3Productivity
If the search target region is restricted to reduce computational load, then the processing speed increases, but the risk of missing valid corresponding points increases
Solution Approach 1:
The patent applies parameter changes by dynamically adjusting the search target region based on image-specific parameters such as occlusion patterns and viewing angles. Rather than using a fixed search region, the system adapts the search boundaries to the specific characteristics of each image pair, optimizing the balance between processing speed and reliability for each case
Data Source
AI summary
An image processing apparatus obtains a first captured image from a first image capturing device and a second captured image from a second image capturing device, and a first image capturing parameter in capturing the first captured image and a second image capturing parameter in capturing the second captured image. Three-dimensional shape data indicating a shape of an object image-captured by the first and second image capturing devices is obtained. Parallax information on the first and second captured images is obtained based on a search point included in the first captured image and a corresponding point corresponding to the search point and included in the second captured image. A search target region limiting a range of a pixel set as the search point is set in the first captured image based on the first and second image capturing parameters and the three-dimensional shape data.