Action Recognition Model Selection Using Principal Components

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing action recognition systems struggle to accurately classify a plurality of complex actions due to reliance on binary determinations, leading to decreased accuracy in intention estimation.

Innovation Solution

An action recognition system utilizing a server-client configuration with a learning function and action recognition function, employing principal component analysis and machine learning to process body frame information and joint angles, and selecting an action classification model based on principal components to enhance accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If binary determination method is used to estimate action intention, then the system is simple to operate, but the accuracy in classifying complex actions of multiple types is markedly decreased

Engineering Contradiction:
Improvesimplicity of intention estimationVSAvoidaccuracy of action classification
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent segments the action recognition process into multiple independent classification models, each specialized for specific action types. Instead of using a single binary determination system, the patent divides the classification task into multiple specialized classifiers that work together to achieve accurate multi-class action recognition while maintaining operational simplicity through modular architecture.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements a dynamic selection mechanism that automatically chooses the appropriate classification model based on the input data characteristics. The system dynamically switches between different classification approaches (including binary determination when appropriate and multi-class classification when needed), allowing it to maintain simplicity for straightforward cases while achieving high accuracy for complex actions.

Inventive Principle:
Principle #15Dynamics

2Measurement precision

If all body frame points are used for action recognition, then the recognition accuracy is improved, but the system becomes vulnerable to occlusion and decreases in reliability

Engineering Contradiction:
Improveaction recognition accuracyVSAvoidrobustness to occlusion
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent extracts and utilizes only the necessary body frame points for specific action types rather than requiring all body frame points. By identifying and extracting the minimal sufficient set of features needed for each action classification, the system maintains high recognition accuracy while becoming more robust to occlusion, as missing points do not critically impact performance.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies different classification strategies to different body frame regions based on their local characteristics and importance. Each action classification model is trained on specific local features relevant to that action type, allowing the system to maintain high accuracy for each local region while being tolerant of occlusion in other areas.

Inventive Principle:
Principle #3Local quality

3Adaptability or versatility

If component analysis is performed to generate statistical components, then the ability to recognize multiple action types is improved, but the device complexity increases

Engineering Contradiction:
Improvecapability to recognize multiple action typesVSAvoidcomplexity of classification system
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent creates a universal component analysis framework that serves multiple classification models simultaneously. The statistical components generated through component analysis are reused across different action classification tasks, allowing the system to recognize multiple action types with a single set of extracted features, thereby reducing overall system complexity while maintaining versatility.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent performs component analysis and generates statistical components in advance, before the actual action classification takes place. This preliminary extraction of features creates a reusable representation that simplifies subsequent classification tasks, as the complex component analysis is performed once and then applied across multiple classification scenarios.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3965007B1Action recognition apparatus, learning apparatus, and action recognition method
Publication Date: 2026.05.06 HITACHI LTD
  • EP3965007B1 patent drawingFigure 1~2
  • EP3965007B1 patent drawingFigure 3
  • EP3965007B1 patent drawingFigure 4

AI summary

An action recognition apparatus performs: a detection process of detecting a shape of an object to be recognized from to-be-analyzed data; a component analysis process of generating one or more components and respective contribution ratios of said components through the component analysis, based on the shape of the object to be recognized; a determination process of determining an ordinal number indicating each dimension of said components based on a cumulative contribution ratio attained from the respective contribution ratios; a selection process of selecting, among the group of action classification models, a specific action classification model in which learning was performed with the same component group as a specific component group including the components with the ordinal number indicating the dimension; and an action recognition process of inputting the specific component group into the specific action classification model, thereby outputting recognition results indicating actions of the object to be recognized.