Assessing the Robustness of Deep Learning-Based Gait Recognition Systems Against Adversarial Attacks

The chapter explores the critical vulnerabilities of deep learning-based gait recognition systems, which have gained prominence for their non-invasive identification capabilities. It delves into the integration of advanced neural network architectures, such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), which have significantly improved the accuracy and reliability of gait recognition. However, these systems are increasingly susceptible to adversarial attacks, where subtle manipulations in input data can lead to incorrect classifications, posing substantial security risks. The chapter introduces a novel method that combines Proximal Policy Optimization (PPO) with Generative Adversarial Networks (GANs) to create and implement adversarial patches, specifically designed to test the resilience of gait recognition models. This innovative approach provides a detailed evaluation of the impact of these patches on challenging the robustness of deep learning-based gait recognition systems. The research methodically assesses the effectiveness of the proposed method against existing techniques, using established gait datasets to empirically showcase its superiority. The chapter also includes a comparative analysis of the system's performance under normal and adversarial conditions, highlighting the need for enhanced defense mechanisms against adversarial threats. The findings underscore the importance of reinforcing gait recognition technologies against evolving cyber threats, making this chapter a crucial read for those interested in the intersection of deep learning and cybersecurity.