Stacked-gait: A human gait recognition scheme based on stacked autoencoders.
Asif Mehmood1, Javeria Amin2, Muhammad Sharif1
1Department of Computer Science, COMSATS University Islamabad, Wah Campus, Wah Cantt, Pakistan.
Plos One
|October 23, 2024
Summary
This study introduces Stacked-Gait, a novel hybrid method for human gait recognition (HGR) that overcomes challenges like clothing variations. The system achieves high accuracy, demonstrating its effectiveness for biometric identification.
Area of Science:
- Computer Science
- Biometrics
- Machine Learning
Background:
- Human gait recognition (HGR) is a biometric technique for individual identification based on walking patterns.
- HGR is valuable for surveillance but faces performance degradation due to variations in clothing, carrying objects, and walking style.
Purpose of the Study:
- To develop a robust hybrid method, Stacked-Gait, for accurate human gait recognition.
- To address the performance limitations of existing HGR methods caused by real-world variations.
Main Methods:
- The Stacked-Gait method involves image resizing and grayscale conversion for feature extraction.
- It utilizes stacked autoencoders for feature learning from training data and extracting features from test data.
- Extracted feature vectors are classified using various machine learning algorithms.
Main Results:
- The Stacked-Gait system was evaluated on the CASIA-B dataset.
- Achieved high accuracy rates across different viewing angles, including 99.90% at 0 degrees and 100% at 90 degrees.
- Demonstrated superior performance compared to recent HGR schemes.
Conclusions:
- Stacked-Gait offers a promising and effective solution for human gait recognition.
- The hybrid approach successfully mitigates challenges posed by variations in gait data.
- The method shows significant potential for enhancing biometric security and surveillance systems.
