Stereo Camera Patch Filtering for Depth Determination Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Stereo depth determination is challenged by differences in blurring between images captured by spatially offset cameras, particularly due to contamination such as dirt on the lenses, which can lead to erroneous matches and incorrect depth determinations.
Innovation Solution
Implementing filtering techniques, such as sharpening or blurring operations, on candidate patches from one camera image to match reference patches from another camera image, with the amount of filtering dependent on the relative difference between the two patches, to improve the accuracy of depth determinations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If images are captured by spatially offset cameras for stereo depth determination, then depth information can be obtained, but differences in blurring between images occur due to contamination such as dirt on lenses
Solution Approach 1:
The patent applies different filtering operations to different regions of the image based on local blur characteristics. Specifically, it identifies regions with contamination-induced blur and applies selective filtering only to those areas, rather than uniformly processing the entire image. This localised approach maintains match accuracy in clean regions while correcting blur in contaminated regions.
Solution Approach 2:
The patent dynamically adjusts filtering parameters based on the detected blur characteristics of each image region. It analyzes the blur程度 in different areas and modifies the filtering strength and type accordingly, changing the processing parameters to match the local image quality conditions and optimize correspondence matching.
2Measurement precision
If filtering operations are applied to candidate patches to compensate for blurring differences, then match accuracy improves, but processing complexity increases
Solution Approach 1:
The patent performs preliminary analysis of blur characteristics in the reference and candidate images before executing the correspondence matching process. By pre-identifying blurred regions and determining appropriate filtering parameters in advance, it avoids complex real-time adjustments during matching, thereby reducing overall processing complexity while maintaining accuracy.
Solution Approach 2:
The patent divides the image processing into distinct segments: blur detection, filtering parameter determination, filtering application, and correspondence matching. This segmentation allows each step to be optimized independently and processed efficiently, reducing the computational complexity compared to a unified approach.
Data Source
AI summary
Methods and apparatus for processing images captured by cameras for use in stereo depth determinations are described and/or for using depth information generated by processing such images is described. Images are captured using a reference camera and at least one additional camera. Filtering is implemented as part of a patch matching process to reduce the risk of erroneous matches, e.g., due to image blur in one image but not another. The filtering may be, and sometimes is, implemented on a patch basis. A candidate patch is generated by filtering a portion of an image captured by a camera, e.g., a second camera by performing a sharpening or blurring operation to a portion of the image to generate a candidate patch. The amount of blurring or sharpening that is applied to generate the candidate patch depends, in some embodiments, on the relative difference between the candidate patch and the reference patch. Thus in some embodiments the amount of sharpening and blurring is dependent on both the content of the reference patch as well as the portion of the second image used to generate the candidate patch.