Recognition of Human Gait Under Asymmetric Loading

Area of Science:

• Biometrics
• Computer Science
• Human Motion Analysis

Background:

• Biometric gait recognition offers non-invasive identification without subject engagement.
• Current methods often neglect gait disruptions, limiting real-world applicability.
• Asymmetric loading is a common factor that can significantly alter a person's gait.

Purpose of the Study:

• To develop and evaluate a robust gait recognition system capable of identifying individuals despite gait disruptions caused by asymmetric loading.
• To explore the effectiveness of ensemble classifiers with deep neural networks for this challenging task.

Main Methods:

• Proposed a solution using ensemble classifiers with various deep neural networks as base classifiers.
• Employed data augmentation techniques to enhance the generalization ability of base models.
• Tested the system on a dataset of 215 individuals (7351 gait cycles) with two decision-combining strategies.

Main Results:

• Achieved high accuracy rates, ranging from 98.55% to 99.8% correct recognitions across different scenarios.
• Demonstrated significantly superior performance compared to existing gait recognition methods in the literature.
• The proposed ensemble approach effectively handles gait variations induced by asymmetric loading.

Conclusions:

• The developed ensemble classifier system shows exceptional performance in biometric gait recognition, even under gait-disrupting conditions.
• This approach offers a promising advancement for reliable person identification in real-world scenarios.
• The findings suggest deep learning-based ensemble methods are highly effective for robust human identification.