Research on Efficient Gait Authentication Techniques Based on Transformer and XGBoost Hybrid Models

This paper addresses the challenges of traditional identity authentication methods by proposing a novel gait-based authentication technology. This paper introduces a lightweight model based on the Transformer architecture, which utilizes the self-attention mechanism to effectively capture global correlations in gait sequences while enabling efficient parallel computation. Our experimental framework employs the HTC Nexus One dataset from McGill University, incorporating innovative data preprocessing techniques such as multi-gait combination sampling and threshold-based gait segmentation to significantly enhance feature representation and recognition reliability. The experimental results demonstrate that our model achieves a remarkable accuracy of 93.82% on the test set, surpassing existing gait recognition methods. To further enhance classification performance, this paper integrates XGBoost as the final classifier, developing a hybrid model that combines Transformer and XGBoost. This hybrid approach achieves an impressive accuracy of 96.17% on the test set, representing a substantial improvement over the standalone Transformer model. Our research not only reduces computational complexity but also offers an innovative solution for identity authentication in high-security scenarios. The proposed methodology shows promising application potential in various fields, including intelligent security systems and smart home technologies, paving the way for more robust and efficient authentication systems.