Multi-Person Pose Grouping for Image-Based Performance Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image analysis techniques fail to effectively handle images containing multiple individuals, lacking the capability to analyze and classify poses of multiple persons captured in a single image.
Innovation Solution
A pose analyzing apparatus and method that estimates poses for each person in an image, classifies them into groups based on their poses, and outputs group information, utilizing machine learning-based models for pose estimation and classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing image analysis techniques are used, then single-person pose analysis is possible, but multi-person pose analysis and classification capability is lacking
Solution Approach 1:
The system segments the image into multiple person instances, estimates poses for each person independently, and then classifies them into different pose groups. This segmentation approach enables multi-person analysis by processing each person separately while maintaining the ability to compare and classify them as a group, thus achieving versatility without overwhelming complexity
Solution Approach 2:
The patent introduces an intermediary classification step that acts as a mediator between pose estimation and final analysis. This intermediary layer groups persons into pose groups based on their estimated poses, enabling the system to handle multiple persons systematically by first organizing them into manageable categories before performing higher-level analysis
2Measurement precision
If pose estimation is performed for multiple persons, then comprehensive pose analysis is achieved, but processing time and computational resources increase
Solution Approach 1:
The system performs preliminary pose estimation for all persons in the image before进行分类 into pose groups. By preparing the pose data for all individuals in advance and organizing them into groups beforehand, the system optimizes the processing flow and reduces overall computation time while maintaining accurate pose analysis for each person
Solution Approach 2:
The patent merges the pose estimation results of multiple persons into unified pose groups. Instead of processing each person completely independently, the system combines their pose data and classifies them into groups, which allows for more efficient processing by handling similar pose patterns together and reducing redundant computations
3Loss of information
If persons are classified into pose groups, then group-level insights are obtained, but individual pose details may be lost
Solution Approach 1:
The system applies local quality by maintaining detailed pose information for each individual person while simultaneously creating group-level classifications. Each person retains their specific pose characteristics (local quality) while being organized into pose groups (global structure), thus preventing information loss at the individual level while enabling efficient group analysis
Solution Approach 2:
The patent adds a new dimension of organization by introducing pose groups as an additional layer of classification. Instead of only analyzing individuals or only grouping them broadly, the system creates intermediate pose groups that provide a middle dimension of analysis, allowing simultaneous attention to individual details and group patterns without losing information in either direction
Data Source
AI summary
A pose analyzing apparatus acquires a target image and estimates a pose for each person. The target image includes two or more persons captured by a camera. The person does arbitrary thing, such as giving a performance, doing exercises, playing music instruments, etc. The pose analyzing apparatus classifies the persons into two or more pose groups based on the poses of the persons, and outputs group information that includes information about at least one of the pose groups.