Gait-based Human Identification through Minimum Gait-phases and Sensors.

Area of Science:

• Biometrics and Human-Computer Interaction
• Gerontology and Healthcare Technology
• Machine Learning and Signal Processing

Background:

• The increasing elderly population necessitates innovative healthcare solutions like Ambient Assisted Living (AAL).
• Contactless human identification is crucial for monitoring and assistive services in AAL environments.
• Gait analysis offers a robust biometric for identification, overcoming limitations of appearance-based methods.

Purpose of the Study:

• To develop and evaluate a gait identification technique using temporal and statistical features from gait phases.
• To assess the accuracy of identification using limited gait cycle information and minimal sensors.
• To compare the performance of various machine learning algorithms for gait-based identification.

Main Methods:

• Gait data collected from 60 individuals using pelvis and foot sensors.
• Features extracted based on temporal and descriptive statistics of different gait phases.
• Six machine learning algorithms (including ANN and SVM) applied for identification tasks.

Main Results:

• High identification accuracy (>95.5%) achieved using a single gait phase and a single sensor.
• 100% identification accuracy attained when monitoring the entire gait cycle with combined pelvis and foot sensors.
• Artificial Neural Networks (ANN) demonstrated superior robustness with fewer data features compared to Support Vector Machines (SVM).

Conclusions:

• Gait identification is feasible and accurate even with partial gait cycle data and minimal sensing.
• The proposed method effectively supports contactless identification for Ambient Assisted Living applications.
• ANN is identified as the optimal machine learning algorithm for gait-based identification in AAL settings.