Related Experiment Video
Updated: Jan 25, 2026

Clinical-oriented Three-dimensional Gait Analysis Method for Evaluating Gait Disorder
Published on: March 4, 2018
Evaluating gait system vulnerabilities through PPO and GAN-generated adversarial attacks
El Mehdi Saoudi1, Jaafar Jaafari2, Said Jai Andaloussi2
1Laboratory of Data Engineering and Intelligent Systems, Faculty of Sciences Ain Chock, Hassan II University of Casablanca, Casablanca, 20000, Morocco. elmehdi.saoudi@univh2c.ma.
Deep learning gait recognition is vulnerable to adversarial attacks. New methods using Generative Adversarial Networks (GANs) and Proximal Policy Optimization (PPO) create patches that fool systems, reducing accuracy and highlighting security risks.
Area of Science:
- Computer Science
- Artificial Intelligence
- Biometrics
Background:
- Deep learning models are increasingly used for gait recognition in security.
- These systems are vulnerable to adversarial attacks, compromising reliability.
- Gait recognition is critical for high-security applications demanding accuracy.
Purpose of the Study:
- To investigate vulnerabilities in deep learning gait recognition systems.
- To develop and deploy adversarial attacks using Generative Adversarial Networks (GANs) and Proximal Policy Optimization (PPO).
- To assess the impact of these attacks on system accuracy and reliability.
Main Methods:
- Integration of Proximal Policy Optimization (PPO) with Generative Adversarial Networks (GANs).
- Creation of targeted adversarial patches designed to deceive gait recognition algorithms.
- Deployment of attacks on CASIA Gait Database: Dataset B and OU-ISIR Treadmill Dataset B.
- Evaluation of attack effectiveness by measuring recognition accuracy decline.
Main Results:
- Adversarial patches effectively deceived gait recognition algorithms.
- Significant decline in recognition accuracy observed after attacks.
- Attacks remained undetected by human oversight.
Conclusions:
- Deep learning gait recognition systems exhibit critical security flaws.
- The developed methodology effectively demonstrates system vulnerabilities.
- Findings inform the development of more robust and secure biometric systems against cyber threats.
Related Concept Videos
Acid Attack on Concrete
The rate at which hydrogen...
Sulfate Attack on Concrete
Sulfates from sources like soil, groundwater, or industrial effluents...
Self-Evaluation: Self-Enhancement and Self-Verification
DC Generator
Nursing Evaluation
Generation Time

