Skeleton Model Estimation for Occluded Limbs via Feature Point Mirroring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image processing technologies face challenges in accurately discriminating subjects captured in images, particularly due to obstruction of body parts, leading to unnatural estimation results.
Innovation Solution
An image processing device and medium that detect and handle feature points from the left and right limbs of a person, estimating hidden body parts by assigning reliability indices and determining feature points as either hidden or non-hidden based on calculated representative values or counts, thereby correcting unnatural estimation results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If skeleton model estimation is applied to images with obscured body parts, then subject discrimination can be performed, but estimation accuracy deteriorates due to hidden body parts not being captured
Solution Approach 1:
The patent creates a virtual copy of the obscured body part by mirroring the visible corresponding body part (e.g., copying the left arm to reconstruct the right arm when obscured). This virtual copy is then integrated into the skeleton model to complete the subject representation despite information loss in the original image
Solution Approach 2:
The patent applies asymmetric handling to left and right body parts by detecting which side is obscured and selectively mirroring only the visible side to reconstruct the hidden side. This asymmetric approach allows the system to adapt to different obscuration scenarios (left or right side blocked) while maintaining accurate skeleton estimation
2Quantity of substance
If feature points from both left and right limbs are detected independently, then detection completeness is improved, but estimation reliability deteriorates when body parts are obscured
Solution Approach 1:
The patent implements a feedback mechanism where the detection results of one limb are used to validate and correct the detection of the other limb. By comparing feature points from both sides and using reliability indices, the system can identify and correct erroneous detections in obscured regions based on the reliable detections from visible regions
Solution Approach 2:
The patent performs preliminary detection of feature points on both limbs before final estimation, allowing the system to identify potential issues early. By calculating reliability indices in advance and comparing results, the system can prepare correction strategies before final skeleton model generation
Data Source
AI summary
An image processing device for improving accuracy of discrimination of hidden portions of a subject captured in an image is disclosed. In this regard the image processing device receives image data of an image in which a person is captured. Based on the received image data, the image processing device estimates a hidden body part of the person that is not captured in the image due to obstruction by another body part of the person. Such estimation is performed by detecting first feature points included in a left limb of the person and second feature points included in a right limb of the person, and estimating at least one of the first feature points and the second feature points are to be included in a hidden body part that is obstructed by another body part of the person or a non-hidden body part.