Motion Recognition Data Augmentation With Deformed Symbol Strings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing motion recognition technologies require large amounts of training data for higher-level motions, which are often unique to specific locations and environments, making it difficult to recognize new motions properly.
Innovation Solution
A data augmentation apparatus transforms motion data into symbol strings, deforms these strings based on similarity and frequency analysis, and generates augmented data to increase the variety of training data for higher-level motion recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large amount of training data is collected for higher-level motions, then motion recognition accuracy is improved, but data preparation difficulty increases
Solution Approach 1:
The patent creates synthetic training data by copying and deforming existing motion data. The deforming unit generates varied motion sequences by applying deformation parameters to original motion trajectories, creating realistic synthetic higher-level motion data that mirrors real-world variations without requiring actual recording of each motion type.
Solution Approach 2:
The patent changes parameters of motion data through the deforming unit, which applies transformation parameters to modify motion trajectories. By adjusting parameters such as motion speed, amplitude, and temporal characteristics, the system generates diverse training data from a limited set of original motions, effectively expanding the training dataset without additional data collection.
2Measurement precision
If training data is collected for location-specific higher-level motions, then recognition accuracy for those motions is improved, but adaptability to new locations deteriorates
Solution Approach 1:
The patent creates a universal motion recognition model that can handle multiple locations and motion types through a single trained system. The deforming unit generates synthetic data that encompasses various motion patterns and environmental conditions, allowing the model to learn generalizable features that work across different locations without requiring location-specific training data for each scenario.
Solution Approach 2:
The patent performs preliminary data augmentation by generating diverse synthetic motion data before the model is trained. The deforming unit pre-processes original motion data to create varied training examples that simulate different locations and conditions, so the model learns these variations in advance and can adapt to new locations without additional training data collection.
3Ease of manufacture
If existing motion data is directly used for training, then data preparation simplicity is maintained, but training data diversity is insufficient
Solution Approach 1:
The patent introduces dynamics to the training data generation process through the deforming unit, which dynamically transforms static motion data into varied sequences. The system applies time-varying deformation parameters to create motion trajectories that exhibit temporal variations, making the training data dynamic and diverse while maintaining simplicity in the data preparation process.
Solution Approach 2:
The patent copies and modifies existing motion data through the deforming unit to create diverse synthetic data. By copying original motion trajectories and applying various deformation parameters, the system generates multiple varied versions of the same motion, significantly increasing training data diversity without requiring complex data collection procedures.
Data Source
AI summary
A motion recognition device of the present disclosure includes a transforming unit that transforms first motion data into a first symbol string including a sequence of symbols; a recognizing unit that recognizes the motion of the first motion data based on the first symbol string; and a deforming unit that generates a third symbol string in which the first symbol string is deformed based on a second symbol string. The second symbol string is a sequence in which second motion data corresponding to the motion recognized in the first motion data is transformed into a sequence of symbols. Consequently, a motion recognition model can be machine-learned using the generated third symbol string for example, and decision making based on the recognized motion can be supported.