Orientation Invariant Gait Matching Using Rotation Matrices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing gait-based authentication methods using inertial sensors face challenges in accurately matching gait data due to variations in device orientation, leading to high error rates, as they often rely on signal magnitudes or frequency domain methods that fail to account for dynamic multi-axis coordinate data effectively.
Innovation Solution
An orientation invariant gait-matching algorithm that includes gait cycle identification, cycle splitting, orientation alignment using the Kabsch algorithm, and gait comparison, which minimizes RMS error between rotated sensor data and a reference gait, allowing for accurate identity authentication despite changes in device orientation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If signal magnitude or frequency domain methods are used for gait matching, then the authentication process is simple, but the accuracy deteriorates due to inability to account for dynamic multi-axis coordinate data
Solution Approach 1:
The patent transforms the gait data from the sensor's body-fixed coordinate system to a global navigation coordinate system using rotation matrices derived from orientation data. This parameter transformation allows the gait characteristics to be represented in a orientation-invariant manner, resolving the contradiction by changing the coordinate system parameters rather than using simple magnitude-based methods.
Solution Approach 2:
The patent introduces orientation angles (roll, pitch, yaw) as additional dimensional parameters to characterize the device orientation. By incorporating these angular dimensions and using them to compute rotation matrices, the system transforms 3D gait data into an orientation-invariant representation, thereby improving accuracy without significantly increasing computational complexity.
2Device complexity
If device orientation variations are not accounted for, then the processing is straightforward, but the authentication reliability deteriorates due to high error rates
Solution Approach 1:
The patent performs orientation alignment as a preliminary step before gait matching by computing rotation matrices from orientation data and applying them to transform gait data into a consistent coordinate system. This preliminary coordinate transformation ensures that subsequent gait comparison is performed on orientation-invariant data, thereby improving reliability without requiring complex adaptive processing during authentication.
3Device complexity
If traditional gait matching methods are used without orientation alignment, then the algorithm is simple, but the measurement precision deteriorates due to orientation sensitivity
Solution Approach 1:
The patent replaces direct mechanical comparison of raw gait data with a mathematical coordinate transformation system. Instead of attempting to mechanically align sensors or physically standardize device orientation, the system uses rotation matrices and coordinate system transformations to mathematically eliminate orientation effects, thereby improving precision while keeping the algorithm computationally efficient.
Data Source
AI summary
Aspects of gait matching are described herein. In one embodiment, a method of gait matching includes identifying a gait cycle timing in a data sequence captured by an inertial sensor, splitting the data sequence into a gait cycle segment, resampling the gait cycle segment, and estimating a rotation matrix for the resampled gait cycle segment. The rotation matrix, in one embodiment, can be estimated by minimizing a root mean squared deviation between the resampled gait cycle segment and a reference gait cycle segment. The method further includes calculating an aligned gait cycle segment from the resampled gait cycle segment using the rotation matrix, and generating a similarity measure between the aligned gait cycle segment and the reference gait cycle segment.