Action Recognition Using Time-Sequence ID Pattern Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing action recognition systems require tedious and time-consuming processes to produce label tables, leading to increased manufacturing costs.
Innovation Solution
An action recognition system that includes a display device and a controller capable of generating classification and ID pattern models through time-division processing of feature vectors, allocating identification IDs to clusters, and outputting recognition information based on changes in identification IDs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If label tables are manually produced for allocating natural-language-like names to partial actions, then action recognition accuracy is improved, but manufacturing cost and time consumption increase
Solution Approach 1:
The system automatically generates label tables by extracting action names from operation manuals and processing them through NLP techniques, enabling the system to serve itself without manual intervention. The controller automatically creates the mapping between action patterns and natural language labels, eliminating the need for tedious manual table production while maintaining recognition accuracy
Solution Approach 2:
The system performs preliminary processing of operation manuals to extract and structure action information before actual action recognition occurs. By pre-processing the manual content to generate candidate action names and their corresponding patterns, the system prepares the label table in advance automatically, reducing both manual workload and manufacturing costs
2Measurement precision
If detailed label tables are created for all partial actions, then recognition precision is improved, but device complexity increases
Solution Approach 1:
The system extracts only the necessary action information from operation manuals, separating essential action patterns from redundant details. By extracting key action names, parameters, and patterns selectively, the system creates a streamlined label table that maintains precision without requiring comprehensive documentation of all possible action variations
Solution Approach 2:
The system transforms unstructured operation manual text into structured action patterns with defined parameters. By converting natural language descriptions into standardized action representations with specific parameters and patterns, the system reduces complexity while preserving recognition precision through consistent parameter-based matching
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An action recognition system includes a controller that computes, in a time division manner, a plurality of feature vectors using as feature quantities information of a machine with respect to running that varies in a time sequence, generates a classification model by classifying the feature vectors into a plurality of clusters and allocating identification IDs to the clusters, generates an ID pattern model by allocating the identification IDs to the feature vectors computed in the time division manner on the basis of the classification model and associating a pattern of the identification IDs that vary in a time sequence according to a predetermined action of the machine with identification information of the predetermined action, generates output information for recognizing the predetermined action on the basis of changes in the identification IDs in the time sequence and the ID pattern model, and controls the display device to output the output information.