Stereo Image Processing with Pixel-Level Disparity Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing devices, such as those disclosed in PTL 1, may produce disparities with significant errors due to varying image conditions, leading to inaccurate object detection and potential safety issues in vehicles.
Innovation Solution
An image processing device incorporating a disparity calculation unit, labeling unit, confidence calculation unit, and disparity correction unit to classify subjects, calculate confidence in disparities, and correct disparities based on pixel-level confidence, using neural networks and stereo matching.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If disparity correction is performed using evaluation functions determined in matching blocks units, then some correction of disparity is achieved, but large errors may still remain depending on image status
Solution Approach 1:
The patent applies local quality by calculating confidence values at the pixel level rather than uniformly across entire matching blocks. Each pixel's disparity correction is based on its own confidence value, allowing different regions of the image to be corrected with appropriate confidence levels. This resolves the contradiction by enabling precise local correction where confidence is high while avoiding unreliable correction where confidence is low, thus improving both disparity accuracy and correction reliability simultaneously.
2Measurement precision
If stereo matching is performed to calculate accurate disparity, then object detection accuracy improves, but the processing complexity increases
Solution Approach 1:
The patent segments the stereo matching process into distinct functional components: initial disparity calculation, confidence calculation for each pixel, and selective correction based on confidence thresholds. This segmentation allows the system to perform accurate stereo matching only where needed (where confidence indicates potential error) rather than uniformly across the entire image, thereby reducing overall processing complexity while maintaining disparity accuracy.
Solution Approach 2:
The patent introduces confidence values as an additional parameter that dynamically controls the correction process. By changing the parameter from a uniform correction approach to a confidence-weighted selective correction approach, the system achieves accurate disparity where needed while reducing unnecessary processing elsewhere, thus balancing accuracy requirements with processing complexity constraints.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
An image processing device capable of calculating an accurate disparity is provided. An image processing device 20 includes: a disparity calculation unit 30 that calculates a disparity between images 111 and 112 taken by a plurality of cameras 11 and 12; a labeling unit 40 that classifies a subject appearing in the images 111 and 112 and that assigns a label to each of pixels of the subject classified, the label corresponding to a type of the subject classified; a confidence calculation unit 50 that calculates confidence in the disparity for each label assigned; and a disparity correction unit 60 that corrects the disparity according to the confidence calculated.