Machine Learning Step Counting for Abnormal Gait
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for step counting fail to accurately account for users with physical disabilities who use crutches or walkers due to gait abnormalities and variable step lengths, leading to inaccurate distance and calorie estimation.
Innovation Solution
A method and system using machine learning models to detect arm and leg swing angles, identify false steps and gait abnormalities, and calculate compensation values to determine a user's step count based on these variations and height.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If conventional step counting methods (peak detection or threshold crossing) are used, then the system is simple and easy to implement, but the step count accuracy deteriorates for users with physical disabilities
Solution Approach 1:
The patent changes the parameters used for step detection from simple accelerometer threshold crossing to multi-parameter analysis including arm swing angle, leg swing angle, gait pattern recognition, and machine learning-based classification. This allows the system to adapt to various gait patterns including those with disabilities, thereby improving measurement precision while maintaining reasonable implementation complexity through modular processing
Solution Approach 2:
The patent replaces simple mechanical threshold-based detection with machine learning models that analyze complex motion patterns. The ML models process accelerometer and gyroscope data to classify gait types and detect steps more accurately, substituting rigid mechanical detection rules with adaptive intelligent algorithms that handle variability in human movement
2Ease of manufacture
If distance estimation is based on manually inputted user height, then the calculation is simple, but the distance estimation accuracy deteriorates for users with variable step lengths
Solution Approach 1:
The patent performs preliminary analysis of the user's gait pattern, arm swing characteristics, and leg swing angles before calculating distance. By pre-characterizing the user's walking style and step length variations through machine learning models, the system establishes personalized parameters that improve distance estimation accuracy without requiring complex real-time calculations during the measurement phase
3Ease of operation
If crutch walking with multiple peaks and time gaps is detected using conventional methods, then the detection process is straightforward, but the step detection accuracy deteriorates due to irregular gait patterns
Solution Approach 1:
The patent implements dynamic adaptation to different gait patterns by using machine learning models that learn from and adjust to individual walking styles. The system dynamically identifies characteristics such as multiple peaks, time gaps, and irregular patterns associated with crutch walking, and adapts its detection thresholds and parameters accordingly, maintaining high accuracy across diverse walking scenarios
Solution Approach 2:
The patent incorporates feedback mechanisms where the machine learning model continuously analyzes detected gait patterns and refines its classification and step detection based on observed characteristics. The system uses feedback from detected peaks, time gaps, and motion patterns to adjust its detection parameters, improving accuracy for irregular gaits through iterative learning and adaptation
Data Source
AI summary
A method includes: determining a motion pattern of a user after an initiation of a walking activity; detecting one or more false steps and a gait abnormality associated with the user, based on the motion pattern; detecting an arm swing angle of the user and a leg swing angle of the user, based on the motion pattern and the walking activity; estimating a first variation in the arm swing angle and a second variation in the leg swing angle in the walking activity; calculating, using a machine learning (ML) model of a plurality of ML models, a compensation value associated with the one or more false steps, based on a combination of the first variation and the second variation, and a height of the user; and determining a step count of the user, based on the calculated compensation value and an initial step count of the user.