Stereo Camera Calibration via Parallax-Guided Feature Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Stereo camera systems face challenges in accurately measuring distance to targets when camera units are misaligned, leading to inaccurate calibrations and potential occlusions in reference regions, which complicates the matching process and reduces the speed and accuracy of distance measurement.
Innovation Solution
An image processing device that performs one-dimensional matching to calculate parallax and two-dimensional matching to extract feature points from captured images, allowing for rapid and accurate calibration of camera units by identifying feature points within a predetermined parallax range, thereby improving alignment and measurement accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional calibration methods are used, then calibration can be performed, but the process is time-consuming and accuracy is reduced when camera units are misaligned
Solution Approach 1:
The calibration process is divided into two independent stages: first, one-dimensional matching is performed to calculate parallax values and identify candidate corresponding points; second, two-dimensional matching is performed only on the candidate points to extract feature points. This segmentation reduces the search space and computational time while maintaining calibration accuracy.
Solution Approach 2:
The patent performs preliminary one-dimensional matching to calculate parallax values before performing the more complex two-dimensional matching. This preliminary action filters and pre-processes the data, identifying candidate corresponding points that are likely to be correct, thereby reducing the computational burden and time required for the final calibration.
2Measurement precision
If reference regions are used for calibration, then calibration can be performed, but occlusions in reference regions complicate the matching process
Solution Approach 1:
The patent extracts and utilizes parallax information calculated from one-dimensional matching to identify candidate corresponding points. By extracting this intermediate information, the system can filter out regions with occlusions or poor matching quality before performing two-dimensional matching, thereby reducing the impact of harmful factors like occlusions on the overall calibration accuracy.
Solution Approach 2:
The patent introduces parallax values as an intermediary parameter that mediates between the two images. This intermediary allows the system to evaluate the quality of corresponding points and filter out problematic regions (such as those with occlusions) before final feature point extraction, thereby reducing the harmful impact of occlusions on matching accuracy.
3Measurement precision
If feature points are extracted using two-dimensional matching, then accurate calibration can be achieved, but the process is computationally intensive and time-consuming
Solution Approach 1:
The patent segments the feature point extraction process into two stages: first, one-dimensional matching to calculate parallax and identify candidate points; second, two-dimensional matching only on the candidate points. This segmentation significantly reduces the computational search space and time while maintaining calibration accuracy.
Solution Approach 2:
Instead of performing two-dimensional matching on all pixels, the patent performs this computationally intensive operation only on a subset of candidate corresponding points identified through one-dimensional matching. This partial action approach maintains sufficient calibration accuracy while dramatically improving processing speed.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Provided is an image processing device including: an input interface 12, which is configured to acquire a first captured image 14L and a second captured image 14R that are captured by a plurality of imaging units 11L and 11R; and a controller 13, which is configured to calculate parallax by performing a one-dimensional matching based on pixel values of the first captured image 14L and pixel values of the second captured image 14R, extracts one or more first feature points from a region in the first captured image 14L that includes continuous pixels having a difference in parallax which is within a predetermined range, extract one or more second feature points corresponding respectively to the first feature points by performing a two-dimensional matching with the first feature points, and calibrate the imaging unit 11L based on positions of the first feature points and positions of the second feature points.