Gait Event Detection Using Adaptive Probability Distributions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current algorithms for detecting gait events using inertial measurement units (IMUs) are limited by the need for rule-based threshold settings, which are not adaptable to varying gait patterns, leading to inaccurate event detection and computation errors.
Innovation Solution
A gait analysis method that determines probability distributions based on consecutive motion data to identify event time points of gait events, using an artificial intelligence model trained on reference data to generate prediction probability distributions and update model parameters for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If rule-based algorithms with fixed thresholds are used for gait event detection, then the detection process is simple and fast, but the detection accuracy decreases when gait patterns vary across different subjects
Solution Approach 1:
The patent transforms fixed threshold parameters into adaptive probability distributions. Instead of using static thresholds that work for average cases, the system models gait event detection using probability distributions (e.g., Gaussian distributions) whose parameters (mean, standard deviation) are learned from training data. This allows the detection thresholds to dynamically adapt to different gait patterns while maintaining computational efficiency through closed-form solutions.
Solution Approach 2:
The patent introduces dynamic adaptability by making the detection thresholds variable rather than fixed. The probability distribution parameters are updated based on subject-specific training data, enabling the system to dynamically adjust to individual gait characteristics. This dynamic approach resolves the contradiction by allowing fast detection (through efficient probability calculations) while achieving high accuracy (through subject-specific adaptation).
2Measurement precision
If subject-specific thresholds are used for each individual, then the detection accuracy improves, but the system complexity and data requirements increase
Solution Approach 1:
The patent implements a feedback mechanism where the system learns from training data and updates its probability distribution parameters accordingly. During training, the system receives feedback in the form of labeled gait events and adjusts its parameters to minimize detection errors. This feedback loop enables accurate subject-specific detection without requiring complex manual configuration, as the system automatically learns optimal parameters from data.
Solution Approach 2:
The system performs self-calibration by automatically learning subject-specific parameters from training data without requiring manual threshold setting for each subject. The algorithm autonomously adapts to individual gait patterns through unsupervised or semi-supervised learning, reducing the need for complex manual configuration while maintaining high detection accuracy. This self-service capability simplifies the overall system complexity.
3Ease of operation
If common thresholds are applied to all subjects, then the system is easy to operate, but detection failures occur for subjects with atypical gait patterns
Solution Approach 1:
The patent creates a universal detection framework based on probability distributions that can handle diverse gait patterns through a single unified model. Instead of requiring separate threshold configurations for different subjects, the system uses a universal probability distribution approach that adapts to individual characteristics through parameter learning. This universal framework maintains ease of operation (single system for all users) while improving reliability through adaptive parameter estimation.
Solution Approach 2:
The system performs preliminary learning during a training phase where it collects and analyzes subject-specific gait data to establish appropriate probability distribution parameters before actual detection begins. This preliminary action enables the system to prepare subject-specific detection parameters in advance, ensuring reliable detection for atypical gait patterns without complicating the operational phase. The training phase separates the complexity from daily use, maintaining ease of operation.
Data Source
AI summary
Disclosed are a gait analysis method, a gait analysis device, and a computer-readable storage medium. The method includes: obtaining consecutive N motion data; determining a plurality of probability distributions based on the N motion data, wherein the probability distributions respectively corresponds to a plurality of gait events; and determining an event time point of each gait event belonging to a specific step according to the plurality of probability distributions.