Stereo Image Disparity Calculation Using Target Region Cropping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods for calculating disparity values in image pairs, especially using binocular cameras, are time-consuming and require excessive computational resources when only specific targets are of interest.
Innovation Solution
The method involves performing target detection on each image to generate smaller to-be-processed images based on detection boxes, cropping these images to focus on the target regions, and calculating disparity values within these reduced images, using thresholding, edge detection, or deep learning algorithms to speed up the process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If disparity calculation is performed on the entire image pair, then comprehensive spatial information is obtained, but calculation time and computational resources increase significantly
Solution Approach 1:
The patent divides the image processing task into two stages: first performing target detection to identify regions containing objects of interest, then performing disparity calculation only on those specific regions. This segmentation approach maintains measurement precision for target areas while significantly reducing calculation time by excluding background regions from intensive processing.
Solution Approach 2:
The patent extracts and isolates target regions from the full image pair using detection boxes, then performs disparity calculation exclusively on these extracted regions. This extraction principle eliminates unnecessary computation on non-target areas while preserving accurate disparity measurements for the regions that matter.
2Measurement precision
If disparity calculation is performed on the entire image pair, then complete scene depth information is obtained, but computational resources are excessively consumed
Solution Approach 1:
The patent segments the image processing task by first identifying target regions through detection, then applying disparity calculation only to those segments. This ensures depth information is obtained for all relevant targets while avoiding wasteful computation on extensive background areas, thereby reducing overall computational resource consumption.
Solution Approach 2:
The patent applies different processing qualities to different regions: high-quality disparity calculation is applied locally to target regions where depth information is needed, while background regions receive minimal or no processing. This local quality approach maintains measurement precision for targets while dramatically reducing total computational resource usage.
3Productivity
If target detection boxes are used to define processing regions, then calculation efficiency is improved, but processing accuracy may be affected by detection errors
Solution Approach 1:
The patent performs target detection as a preliminary action before disparity calculation, using detection boxes to pre-identify regions of interest. This preliminary step enables efficient region-based processing while the subsequent disparity calculation on these predefined regions maintains accuracy by focusing computational effort on verified target areas rather than the entire image.
Data Source
Figure 1~2
Figure 3A~3B
Figure 4~5
AI summary
The present disclosure provides an image processing apparatus, an image processing device and a computer-readable storage medium. The image processing method, comprising: performing a target detection on a first image to generate a first to-be-processed image, the first to-be-processed image has a size smaller than the size of the first image; performing the target detection on a second image to generate a second to-be-processed image, the second to-be-processed image has a size smaller than the size of the second image, the first image and the second image constitute an image pair containing the target, and a disparity exists between the first image and second image; calculating, based on the first to-be-processed image and the second to-be-processed image, a disparity value of the target in the first to-be-processed image and the second to-be-processed image.