Motion Data Generation Synchronizing Augmented Datasets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing motion data generation methods struggle with synchronizing motion data across different datasets with varying backgrounds and timing, leading to decreased accuracy in data augmentation, especially when the number of training data is limited.
Innovation Solution
A motion data generation device and method that acquires and groups motion data by class, sets reference data, and synchronizes data to be converted with reference to motion timing, using a data conversion unit that includes a posture estimation unit, feature amount calculation, and synchronization processing to generate extension motion data aligned in the time axis direction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If temporal synchronization is performed using background-influenced methods (NPL 1), then motion data can be synchronized, but accuracy decreases when backgrounds are greatly different between datasets
Solution Approach 1:
The patent extracts and removes the background component from motion data, processing only the motion-specific features. By separating motion data from background information and processing them independently, the system avoids background interference while maintaining synchronization accuracy across different datasets.
Solution Approach 2:
The patent introduces a background removal processing unit as an intermediary component that preprocesses motion data before synchronization. This intermediary layer eliminates background variations, allowing the synchronization algorithm to focus solely on motion features and achieve consistent accuracy across diverse backgrounds.
2Reliability
If spatial augmentation methods (NPL 2) are used, then data robustness to noise is improved, but temporal synchronization capability is lost
Solution Approach 1:
The patent merges spatial augmentation techniques with temporal synchronization processing in an integrated system. By combining the noise-robust spatial transformation capabilities with temporal alignment functionality, the system achieves both reliability against noise and preservation of temporal relationships in augmented data.
3Quantity of substance
If motion data augmentation is performed without temporal synchronization, then data quantity increases, but applicability to tests requiring synchronized timing decreases
Solution Approach 1:
The patent maintains continuous temporal synchronization throughout the data augmentation process. By ensuring that the temporal alignment relationship is preserved during augmentation operations, the system generates large quantities of usable data that remain applicable to tests requiring synchronized timing, preventing loss of temporal information.
Data Source
AI summary
Provided is a motion data generation device including an acquisition unit that acquires a plurality of pieces of motion data to be data converted, a conversion data choosing unit that groups the plurality of pieces of motion data for each motion class that is a target motion for data augmentation, a data conversion unit that sets at least one piece of the motion data grouped for the each motion class for reference data, sets at least one piece of the motion data different from the reference data among the grouped motion data for data to be converted, and generates extension motion data in which the data to be converted is synchronized with reference to motion timing of the reference data, and an output unit that outputs the generated extension motion data.