Action Recognition Model Learning via Error-Added Movement Locus Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for recognizing human actions from movement loci captured by cameras suffer from low recognition accuracy due to estimation errors, leading to misclassification of actions like 'normal walking' and 'wandering', and potential clustering of distinct features.
Innovation Solution
A model learning device and method that generates error-added movement locus data by incorporating estimation errors into the learning process, using this data to train an action recognition model that can accurately distinguish between different actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If movement locus data is used directly for action recognition learning, then the learning process is simple, but recognition accuracy is low due to estimation errors
Solution Approach 1:
The patent applies preliminary action by pre-processing the movement locus data to add estimation errors before the action recognition learning. The error-added movement locus generation unit creates training data that includes simulated estimation errors, so the model learns to recognize actions despite these errors. This preliminary preparation of training data with errors resolves the contradiction by improving recognition accuracy through error tolerance while keeping the learning process structured and manageable.
Solution Approach 2:
The patent changes the parameters of the training data by adding estimation errors to the movement locus data. Instead of using clean, accurate movement locus data directly, the system transforms the data by introducing controlled errors in position and timing parameters. This parameter transformation allows the model to learn robust action recognition that is insensitive to estimation errors, thereby improving recognition accuracy without requiring overly complex learning algorithms.
2Measurement precision
If estimation errors are added to movement locus data, then recognition accuracy improves, but data processing complexity increases
Solution Approach 1:
The patent extracts and separates the error addition process from the main action recognition learning process. The error-added movement locus generation unit is a dedicated component that specifically handles error injection, while the action recognition model learning unit focuses on learning from the prepared data. This separation makes the data processing difficulty manageable by isolating the complex error simulation task to a specialized module, allowing the rest of the system to remain relatively simple.
3Reliability
If clean movement locus data is used for training, then the model learns ideal patterns, but misrecognition occurs in real-world conditions with estimation errors
Solution Approach 1:
The patent converts the harmful effect of estimation errors into a beneficial training mechanism. Instead of treating estimation errors as noise to be eliminated, the system deliberately adds these errors to training data, transforming them into a tool for improving model robustness. The error-added movement locus generation unit simulates real-world estimation errors during training, allowing the model to learn patterns that are reliable under actual operating conditions. This approach converts the previously harmful estimation errors into a benefit that enhances action recognition reliability.
Data Source
AI summary
A model learning device provided with: an error-added movement locus generation unit for adding an error to movement locus data for action learning that represents the movement locus of a subject and to which is assigned an action label that is information representing the action of the subject, and thereby generating error-added movement locus data; and an action recognition model learning unit for learning a model, using at least the error-added movement locus data and learning data created on the basis of the action label, by which model the action of some subject can be recognized from the movement locus of the subject. Thus, it is possible to provide a model by which the action of a subject can be recognized with high accuracy on the basis of the movement locus of the subject estimated using a camera image.