Body Hand Association via Bounding Box Occlusion Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current body and hand association methods in complex scenes with many people face challenges in accuracy due to occlusions and crossed arms, relying on Human Keypoints Detection (HKD) which is costly and prone to errors when determining hand ownership.
Innovation Solution
A method that acquires images, determines body and hand bounding boxes, calculates association probabilities, and forms circumscribed boxes to assess the degree of association based on hand key points, improving accuracy by using a combination of feature extraction and association algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Human Keypoints Detection (HKD) is used to determine hand ownership, then hand key points can be detected, but judgment accuracy is poor and judgment cost is high
Solution Approach 1:
The patent segments the hand ownership determination process into multiple independent features: association probability between body and hand bounding boxes, occlusion relationships, and spatial positional relationships. Each feature is calculated separately and then integrated, avoiding the need for complex HKD while improving accuracy through multi-factor analysis
Solution Approach 2:
The patent performs preliminary calculations of association probabilities and occlusion relationships between body and hand bounding boxes before final hand ownership determination. This preliminary analysis provides a foundation for accurate judgment without requiring costly HKD, as the association metrics are computed in advance using simpler image processing techniques
2Measurement precision
If HKD is used in complex scenes with occlusions and crossed arms, then hand key points can be identified, but accuracy deteriorates due to occlusions
Solution Approach 1:
The patent introduces an intermediary occlusion detection mechanism that analyzes the spatial relationships and visibility between body and hand bounding boxes. By detecting occlusion states as an intermediate step, the system can adjust its hand ownership determination to account for occlusions, thereby maintaining accuracy in complex scenes where direct HKD fails
Solution Approach 2:
The patent moves from 2D key point detection to 3D spatial relationship analysis by considering depth information through occlusion detection and positional relationships. This dimensional transition allows the system to disambiguate crossed arms and occluded hands by analyzing their spatial configurations rather than relying solely on 2D key point coordinates
Data Source
AI summary
A body and hand association method includes: an image to be detected of which an image content includes a body and a hand is acquired; a body bounding box of the body and a hand bounding box of the hand are determined in the image to be detected; an association probability between the body bounding box and the hand bounding box is determined; a circumscribed box of the body bounding box and the hand bounding box is determined; and a degree of association between the body and hand in the circumscribed box is determined based on a hand key point in the circumscribed box and the association probability.