Accelerometer-Based Gait Recognition by Sparse Representation of Signature Points With Clusters
IEEE Transactions on Cybernetics
|November 26, 2014
Summary
This study introduces a new gait recognition algorithm using signature points (SPs) extracted from acceleration signals. This method improves accuracy and avoids issues found in previous step-cycle-based approaches for human identification.
Area of Science:
- Biometrics
- Signal Processing
- Machine Learning
Background:
- Gait is a promising biometric for human recognition, captured via acceleration signals from smart devices.
- Existing methods often fail due to step-cycle detection issues and phase misalignment.
- Wearable sensors offer non-intrusive gait data collection for applications like access control.
Purpose of the Study:
- To propose a novel gait recognition algorithm that overcomes limitations of existing methods.
- To enhance accuracy and robustness in accelerometer-based human identification.
- To develop a system that does not rely on explicit step-cycle detection.
Main Methods:
- A multiscale signature point (SP) extraction method, including localization and descriptor generation.
- A sparse representation scheme encoding new SPs using known ones, leveraging phase propinquity for physical meaningfulness.
- A classifier for sparse-code collections associated with gait signal series.
Main Results:
- The proposed algorithm outperformed existing methods, even when step cycles were perfectly detected.
- Achieved 95.8% rank-1 accuracy for identification using accelerometers from five body locations.
- Attained an equal error rate of 2.2% for verification.
Conclusions:
- The novel SP-based algorithm offers a robust and accurate solution for gait recognition.
- Eliminates reliance on step-cycle detection, addressing a key limitation in prior work.
- Demonstrates high performance for both human identification and verification tasks.
