Vertical Medial Axis Detection for Robust Object Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods for object matching across different camera views face challenges such as requiring costly foreground detection, being sensitive to posture variations, and not providing confidence measures, especially when dealing with non-rigid objects and varying lighting conditions.
Innovation Solution
A method that determines a continuous vertical medial axis of an object by calculating horizontal symmetry for each pixel and establishing first and second vertical bounds within the image, allowing for robust object matching without the need for foreground detection and accommodating posture variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If appearance-based object matching is performed using foreground mask and vertical symmetry axes, then object matching capability is improved, but computational cost and memory usage increase significantly
Solution Approach 1:
The patent extracts only the essential symmetry information needed for matching by determining a single vertical symmetry axis for the entire object rather than processing the complete foreground mask. This extraction approach retrieves the critical symmetry feature (vertical axis location) while discarding redundant computational overhead associated with full mask processing, thereby reducing computational cost while preserving matching capability.
Solution Approach 2:
Instead of first segmenting the foreground object into a mask and then finding symmetry axes (traditional approach), the patent inverts the process by directly determining the vertical symmetry axis from the bounding box of detected object points without requiring explicit foreground mask generation. This inversion eliminates the computationally expensive mask processing step while maintaining the essential symmetry information needed for matching.
2Measurement precision
If vertical symmetry axes are determined from foreground object mask, then symmetry features are obtained, but the method fails when objects are tilted or foreshortened
Solution Approach 1:
The patent applies asymmetry principle by using an asymmetric bounding box (rectangular bounding box) that adapts to the object's orientation in the image plane. Unlike methods that assume upright posture and use symmetric coordinate systems, this approach allows the bounding box to be rotated and skewed to match the object's actual orientation, enabling accurate symmetry axis determination for tilted or foreshortened objects while maintaining measurement precision.
3Manufacturing precision
If foreground mask is determined by segmentation, then object boundaries are identified, but the process is computationally expensive and memory intensive
Solution Approach 1:
The patent performs preliminary action by directly computing the bounding box from detected object points before any symmetry axis determination. This preliminary bounding box computation serves as a simplified precursor that captures the essential spatial extent and orientation of the object without requiring complex segmentation. By establishing the bounding box first, the method prepares the necessary geometric framework while avoiding the computational burden of full foreground mask segmentation.
Data Source
AI summary
A method of determining a vertical medial axis of an object in an input image, is disclosed. A measure of horizontal symmetry is determined for each pixel of the image. First and second vertical bounds within the image are determined. A continuous path between the first and second vertical bounds is determined based on the horizontal symmetry measure, the continuous path representing the vertical medial axis of the object.