Binocular Image Matching via Bounding Box Regression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing binocular image matching algorithms are complex and have high error rates due to the use of various features and threshold values, affecting the robustness and accuracy of target matching.
Innovation Solution
Perform target detection on a first image to obtain a first bounding box, determine a second bounding box in a second image without additional detection, and regress a third bounding box using feature extraction and regression networks to achieve accurate matching.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional matching algorithms with multiple features and thresholds are used, then target matching can be performed, but the algorithm complexity increases and matching accuracy decreases
Solution Approach 1:
The patent extracts and removes the complex matching algorithm portion from the traditional binocular vision system. Instead of using multiple features and thresholds for matching, the invention directly uses the detected bounding box from the first image to locate the corresponding target in the second image through coordinate transformation, eliminating the need for complex matching algorithms while improving both accuracy and simplifying the system.
Solution Approach 2:
The patent inverts the traditional approach by not trying to match features between images to find corresponding targets. Instead, it directly transforms the bounding box coordinates from the first image to the second image using pre-calibrated camera parameters, reversing the conventional logic of feature-based matching and achieving simpler, more accurate results.
2Reliability
If target detection is performed on both binocular images followed by matching algorithm, then complete target information is obtained, but calculation overhead increases significantly
Solution Approach 1:
The patent performs target detection only on the first image beforehand, obtaining the bounding box in advance. The bounding box coordinates are then directly transformed to the second image using pre-calibrated camera parameters, avoiding the need to perform detection on both images. This preliminary action approach reduces calculation overhead while maintaining target matching robustness.
Solution Approach 2:
The patent creates a transformed copy of the bounding box from the first image and applies it to the second image through coordinate transformation. Instead of independently detecting targets in both images and then matching them, the invention copies the detection result from the first image and transforms it to the second image coordinate system, significantly reducing computational requirements.
Data Source
AI summary
Embodiments of the present invention disclose a binocular image matching method, apparatus, device, and storage medium. The method comprises: performing target detection on a first image to obtain a first bounding box of a target in the first image; determining a second bounding box corresponding to the first bounding box in a second image; and obtaining a third bounding box of the target in the second image by regressing the second bounding box. The technical solutions realize accurate matching of targets in a binocular image without requiring performing target detection on both images in the binocular image, and then use a matching algorithm to match the targets detected in the two images, thereby greatly reducing the calculation overhead of target matching in the binocular image.