Person Identification Device Contour Feature Weighting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image determination methods struggle to accurately identify whether individuals captured from different directions in surveillance images are the same person, particularly when clothing has open fronts, leading to significant differences in feature values between front and back images.
Innovation Solution
The proposed solution involves an image determination device that extracts contour color information from images captured from different angles, weighting the comparison results of contour features more heavily than other features, allowing for precise identification of individuals regardless of direction, by using a processing device with a person detection/tracking unit, feature extraction unit, silhouette extraction unit, and same person determination unit.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image determination methods compare feature values of persons from different directions, then identification can be performed, but accuracy deteriorates when clothing has open fronts causing significant differences between front and back images
Solution Approach 1:
The patent segments the person's body into multiple regions (head, upper body, lower body) and further divides each region into contour portions and inner portions. By comparing only the contour portions which are less affected by clothing open fronts, the method achieves more reliable identification across different directions while maintaining measurement precision.
Solution Approach 2:
The patent applies different comparison strategies to different parts of the person's body. Contour portions are given higher weight in comparison because they maintain consistency across front and back views, while inner portions are weighted less. This local quality differentiation resolves the contradiction by focusing comparison on stable features.
2Adaptability or versatility
If feature values from all body portions are compared equally, then comprehensive comparison is achieved, but precision deteriorates due to significant differences in clothing patterns between front and back
Solution Approach 1:
The patent assigns different weights to different body portions based on their stability across directions. Contour portions receive higher weights because they show consistent features, while inner portions with open clothing receive lower weights. This resolves the contradiction by maintaining comprehensive comparison while improving precision through selective weighting.
Solution Approach 2:
The patent introduces asymmetric weighting where contour portions are weighted differently from inner portions. This asymmetric treatment of body regions addresses the symmetry problem of front-back clothing appearances, allowing comprehensive comparison while achieving precision through differentiated weight assignment.
3Loss of information
If clothing features including color and pattern are extracted for identification, then detailed comparison is possible, but reliability deteriorates when clothing has open fronts creating inconsistent features
Solution Approach 1:
The patent segments clothing features into contour region features and inner region features. By extracting and comparing features primarily from contour regions (which remain consistent in front and back views), the method retains sufficient feature detail for identification while avoiding the unreliable inner region features affected by open fronts.
Solution Approach 2:
The patent extracts and isolates contour portion features from the overall clothing features for primary comparison. This extraction of stable contour features separates them from the unreliable inner features, maintaining feature detail where available while improving reliability by excluding inconsistent features.
Data Source
AI summary
A device that determines whether a first person extracted from a first image is identified with a second person extracted from a second image, includes a processor configured to extract a first feature value of a contour portion of the first person, from the first image, extract a second feature value of a contour portion of the second person, from the second image, and determine whether the first person is identified with the second person, based on comparison processing in which a first effect on a first comparison result of the first feature value and the second feature value becomes greater than a second effect on a second comparison result of a third feature value of a portion further from a contour than the contour portion of the first person and a fourth feature value of a portion further from the contour than the contour portion of the second person.