Pose Class Generation for Object Function Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing technologies require manual definition of pose information by users, making it difficult to accurately identify object functions, especially when unknown poses are encountered, and necessitate significant labor for registering pose information.
Innovation Solution
An image processing apparatus that automatically generates and identifies pose classes for a person with respect to an object by extracting and clustering part information from multiple images, allowing for the classification of poses without pre-defined pose information, and associates these classes with the object for function identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual definition of pose information is used, then pose information can be registered in advance, but it requires considerable labor and subjective errors
Solution Approach 1:
The system performs automatic pose information registration through self-learning mechanisms. The learning unit automatically acquires pose information by analyzing images of persons interacting with objects, eliminating the need for manual definition and registration by users. This self-service approach resolves the contradiction by making the system autonomous in acquiring pose data.
Solution Approach 2:
The system performs preliminary pose information acquisition through automatic learning before actual function identification tasks. By pre-acquiring and storing pose information through automated image analysis, the system prepares necessary data in advance without requiring manual intervention during operation, thus reducing both time and labor investment.
2Adaptability or versatility
If pre-registered pose information is used, then object function identification can be performed, but unknown poses cannot be identified
Solution Approach 1:
The system transitions from static pre-registered pose information to dynamic automatic learning. The learning unit continuously acquires new pose information from images, allowing the system to adapt to unknown poses dynamically. This dynamic approach enables the system to handle both known and unknown poses reliably by learning new poses as they appear.
Solution Approach 2:
The system implements feedback mechanisms where the learning unit continuously receives image data, analyzes poses, and updates pose information databases. This feedback loop enables the system to learn from actual usage scenarios, improving its ability to identify both known and unknown poses through continuous refinement based on real-world data.
3Adaptability or versatility
If enormous amount of pose information is registered in advance, then comprehensive pose coverage is achieved, but it requires considerable labor
Solution Approach 1:
The system achieves comprehensive pose coverage through self-service automatic learning rather than manual registration. The learning unit autonomously analyzes images to extract pose information, eliminating the need for labor-intensive manual definition while achieving comprehensive coverage of various poses naturally encountered in usage scenarios.
Solution Approach 2:
The system creates copies of pose information by analyzing and replicating poses from image data. Instead of manually defining each pose, the system automatically copies pose characteristics from observed images, achieving comprehensive pose coverage through automated replication of real-world pose data.
Data Source
AI summary
Image processing apparatus programmed to: continuously shoot a subject to obtain images, and detect the object and extract a position of the object from a three-dimensional position of the subject in the images; detect the person and extract a position of the person from the three-dimensional position, and extract, from the position of the person, part information pieces including respective positions of characteristic parts of the person; generate a pose class for each set of part information pieces, the part information pieces being similar to one another in correlation between parts of the person calculated from each of the part information pieces; identify a pose class to which the correlation between parts of the person belongs, among generated pose classes, when a distance between the person and the object is within a predetermined range; and store the identified pose class in association with the object.