Model-based and model-free deep features fusion for high performed human gait recognition.
Reem N Yousef1, Abeer T Khalil2, Ahmed S Samra2
1Delta Higher Institute for Engineering and Technology, Mansoura, Egypt.
Summary
This study introduces a novel deep convolutional neural network (CNN) for non-contact human authentication using gait recognition. The model achieves high accuracy, offering a secure and efficient biometric solution.
Area of Science:
- Computer Science
- Biometrics
- Artificial Intelligence
Background:
- Increased demand for non-contact biometric systems post-COVID-19 pandemic.
- Need for secure and accurate human authentication methods.
- Limitations of existing biometric models in real-world scenarios.
Purpose of the Study:
- To develop a novel deep convolutional neural network (CNN) for non-contact human authentication.
- To enhance recognition accuracy and efficiency using gait and pose analysis.
- To fuse model-free and model-based feature extraction techniques.
Main Methods:
- Utilized a novel deep convolutional neural network (CNN) model.
- Employed concatenated fusion of CNN with a fully connected model.
- Extracted features from human silhouette images (model-free) and joint data (model-based).
- Tested the system on the CASIA gait dataset (casia-A and casia-B).
Main Results:
- Achieved high accuracy rates of 99.8% on casia (B) and 99.6% on casia (A).
- Demonstrated superior performance compared to state-of-the-art studies.
- Showcased robustness under various covariate conditions.
- Evaluated multiple performance metrics including accuracy, specificity, and sensitivity.
Conclusions:
- The proposed CNN model offers a highly accurate and efficient solution for non-contact human authentication.
- The fusion of model-free and model-based features enhances gait recognition capabilities.
- The system provides a robust real-time authentication method suitable for diverse conditions.


