Action Recognition ID Pattern Modeling for Automated Labeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing action recognition systems require tedious and time-consuming table production for allocating natural-language-like names to partial actions, 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, allowing for efficient recognition and reduced manufacturing costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If manual table production is used for allocating natural-language-like names to partial actions, then action recognition accuracy is maintained, but manufacturing cost and time consumption increase
Solution Approach 1:
The system automatically generates the mapping table between partial actions and natural-language-like names through self-learning from sensor data without requiring manual production. The controller extracts feature quantities from sensor data, determines partial actions through pattern recognition, and automatically creates the mapping relationships, enabling the system to serve itself in generating the previously manually-produced table.
Solution Approach 2:
The patent replaces the manual mechanical process of table production with an automated information processing system. Instead of human operators manually creating the mapping table, the controller uses pattern recognition algorithms to automatically generate the mapping between partial actions and natural-language-like names from sensor data, substituting manual labor with computational processing.
2Ease of manufacture
If automated ID allocation is implemented, then manufacturing cost is reduced, but system complexity increases
Solution Approach 1:
The controller performs multiple functions including sensor data acquisition, feature quantity extraction, pattern recognition for partial action determination, ID allocation, and natural-language-like name assignment. By consolidating these functions into a single controller, the system reduces the need for separate manual processes and external tools, thereby reducing manufacturing cost while managing complexity through functional integration.
Solution Approach 2:
The system transforms the approach from fixed manual table production to dynamic automated ID allocation based on pattern recognition. The controller changes the parameter of action identification from static natural-language-like names to automated ID assignment based on extracted feature quantities and recognized patterns, reducing manufacturing complexity while maintaining recognition accuracy.
Data Source
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.