Pose-Based Image Selection with Missing Keypoint Handling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In image selection processes, especially in surveillance systems, the accuracy of selecting images of a desired pose can be compromised when parts of the human body are hidden or lacking, leading to incorrect selections.
Innovation Solution
An image selection apparatus and method that utilizes query information including relative positions of keypoints to select target images, considering the kind and number of lacking keypoints, ensuring accurate selection even when partial body information is missing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If image selection is performed using pose information from surveillance images, then pose search capability is improved, but selection accuracy deteriorates when parts of the human body are hidden or lacking
Solution Approach 1:
The patent applies preliminary action by complementing lacking keypoint information before performing image selection. A learning model is used to generate estimated keypoint information for body parts that are hidden or lacking in the query image. This preliminary completion of information ensures that subsequent pose-based image selection can proceed with complete data, preventing accuracy degradation due to missing body part information.
2Difficulty of detecting and measuring
If keypoint information is used for image selection, then pose detection capability is improved, but reliability deteriorates when keypoints are lacking due to body concealment
Solution Approach 1:
The system performs preliminary keypoint completion using a learning model before image selection. When the query image contains hidden or lacking body parts, the learning model estimates and supplements the missing keypoint information. This ensures that the pose detection process receives complete keypoint data, maintaining reliable image selection even when the original query image has concealed body parts.
3Productivity
If images with partial body information are included in selection, then search coverage is improved, but selection accuracy deteriorates due to incomplete human body information
Solution Approach 1:
The patent applies preliminary action by using a learning model to complete missing keypoint information before image selection. Images with partially concealed bodies are first processed through the learning model, which generates estimated keypoint positions for hidden body parts. This preliminary completion allows these images to be included in the search without compromising selection accuracy, as the missing information has been restored.
Data Source
AI summary
A search unit (105) of an image selection apparatus (10) includes a query acquisition unit (610) and an image selection unit (620). The query acquisition unit (610) acquires query information indicating a pose of a person. The image selection unit (620) selects at least one target image from a plurality of images subject to selection by using the query information. Herein, the query information includes relative positions of a plurality of keypoints indicating different portions of a human body from each other. Then, the image selection unit (620) selects the at least one target image by using a relative position of the keypoint of a person included in the image subject to selection, at least one of a kind of a lacking keypoint and a number of lacking keypoints, and the query information, the lacking keypoint being the keypoint that lacks in the person.