Synthetic Gesture Data Generation for Limited Training Sets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current gesture recognition technologies face challenges in achieving high accuracy with limited training data, requiring domain-specific knowledge, and being unsuitable for rapid prototyping, especially when dealing with dynamic gestures and continuous data streams.
Innovation Solution
The method involves stochastic resampling and normalization of gesture paths to generate synthetic variants, which can be used to train gesture recognizers with minimal computational overhead and achieve high accuracy, while being modality-agnostic and easily implementable.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional gesture recognition methods (SVM, HMM, neural networks) are used, then recognition accuracy can reach above 90%, but copious training data and advanced machine learning knowledge are required
Solution Approach 1:
The patent applies synthetic data generation by creating artificial gesture samples that replicate the statistical properties of real gestures. Instead of collecting copious real training data, the system generates synthetic gesture trajectories by perturbing and transforming a small set of real gesture samples, thereby copying the essential characteristics needed for training machine learning models without requiring large amounts of actual user data
Solution Approach 2:
The patent transforms gesture data by applying parameter changes such as temporal scaling, spatial transformation, and velocity modification to generate varied synthetic samples. These parameter transformations allow a single real gesture sample to produce multiple diverse training examples, effectively increasing the training data volume without additional user input
2Measurement precision
If traditional gesture recognition methods are used, then recognition accuracy can reach above 90%, but domain-specific knowledge for feature extraction is required
Solution Approach 1:
The patent copies the statistical distribution and temporal-spatial characteristics of real gestures directly into synthetic samples without requiring manual feature engineering. By preserving the raw gesture trajectories and their inherent features through synthetic generation, the system avoids complex domain-specific feature extraction while maintaining the information needed for accurate recognition
Solution Approach 2:
The patent creates a universal synthetic data generation framework that works across different gesture types, devices, and modalities without requiring domain-specific customization. The same synthetic generation pipeline can handle diverse gesture datasets, making the system universally applicable without needing specialized feature extraction knowledge for each domain
3Ease of manufacture
If synthetic data generation is used to reduce training data requirements, then ease of prototyping improves, but generation speed and computational overhead must be minimized
Solution Approach 1:
The patent applies partial action by generating only the essential variations needed for effective prototyping rather than exhaustively covering all possible gesture variations. The synthetic generation focuses on the most impactful transformations (temporal scaling, spatial transformation) that provide sufficient diversity for model training without unnecessary computational overhead from exhaustive sampling
Data Source
AI summary
A method of generating synthetic data from time series data, such as from handwritten characters, words, sentences, mathematics, and sketches that are drawn with a stylus on an interactive display or with a finger on a touch device. This computationally efficient method is able to generate realistic variations of a given sample. In a handwriting or sketch recognition context, synthetic data is generated from real data in order to train recognizers and thus improve recognition accuracy when only a limited number of samples are available. Similarly, synthetic data can also be used to test and validate such recognizers. Also discussed is a dynamic time warping based approach for both segmented and continuous data that is designed to be a robust, go-to method for gesture recognition across a variety of modalities using only limited training samples.