Body-Part Action Determination With Adaptive Model Selection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing pose estimation technologies struggle with accurate detection of body parts in crowded scenes, leading to decreased inference accuracy and action determination reliability due to undetected body parts or low detection reliability.

Innovation Solution

An action determination apparatus and method that employs a detection unit to identify body parts, a selection unit to choose a determination method based on the detected body parts' positional relationships, and a control unit to determine actions using the selected method, adapting to the detection results to improve accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If a machine learning model is trained based on all body parts targeted for detection, then the model can perform comprehensive action determination, but the accuracy decreases when some body parts are not detected or have low detection reliability

Engineering Contradiction:

Improvecomprehensive action determination capabilityVSAvoidaction determination accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The system dynamically adjusts the set of body parts used for action determination based on detection reliability. When certain body parts are not detected or have low reliability, the system automatically adapts by selecting alternative body parts or determination methods, transforming a static model into a dynamic one that responds to detection conditions.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system changes the parameters of the pose estimation model based on detection results. When detection reliability is low for certain body parts, the system modifies which body parts are included in the action determination process, effectively changing the model's input parameters to maintain accuracy.

Inventive Principle:

Principle #35Parameter changes

2Ease of operation

If body parts are used for action determination even when detection reliability is low, then the system can maintain a consistent determination process, but the action determination accuracy decreases

Engineering Contradiction:

Improvedetermination process consistencyVSAvoidaction determination reliability

Core Design Contradiction:

Ease of operationVSReliability

Solution Approach 1:

The system uses detection reliability as feedback to adjust the action determination process. When detection reliability is low, the system receives feedback about the poor quality data and responds by selecting alternative determination methods or body parts, creating a closed-loop system that maintains reliability.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system prepares multiple determination methods in advance to cushion against potential detection failures. By having alternative methods ready before detection issues occur, the system can switch to more reliable methods when detection reliability is low, preventing accuracy degradation.

Inventive Principle:

Principle #11Beforehand cushioning (Prior cushioning)

3Measurement precision

If geometric relationship parameters are adjusted to suppress pose estimation accuracy decrease, then pose estimation accuracy is maintained, but the system cannot handle cases where body parts are not detected

Engineering Contradiction:

Improvepose estimation accuracyVSAvoidhandling of undetected body parts

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The system segments the action determination process into multiple independent determination methods, each using different subsets of body parts. This allows the system to select and apply only the determination methods that are compatible with the detected body parts, effectively handling cases where certain body parts are not detected.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system creates a universal action determination framework that can handle multiple scenarios including complete body part detection, partial detection, and low reliability detection. By designing determination methods that can universally adapt to different detection conditions, the system maintains both accuracy and versatility.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS12462611B2Action determination apparatus, image capturing apparatus, action determination method, and storage medium that detect N types of body parts (N>3)

Publication Date: 2025.11.04 CANON KK

AI summary

A determination control apparatus that functions as an action determination apparatus including a detection unit performing detection processing for detecting N types of body parts (N≥3), the body parts including a first type, of a subject in an image, and a selection unit configured to select one of a plurality of machine learning models for determining an action of the subject, based on a result of the detection processing. The plurality of learning models includes a first machine learning model to perform inference relating to the action based on a positional relationship of the N types of body parts, including the first type of body part and a second model to perform inference relating to the action based on a positional relationship of M types of body parts (N>M≥2) not including the first type of body part, among the N types of body parts.