A Framework for Gait based user Verification using CNN and Transformers Hybrid Model

Our gait-based user verification system provides a practical and non-intrusive way to identify people, even under challenging conditions like changes in clothing or different camera angles. We introduce a unique hybrid model that combines CNNs, GRUs, and Transformers. The CNN extracts spatial features, the GRU captures temporal motion patterns, and the Transformer models long-range dependencies in the gait sequence. By training this model with Triplet Margin Loss, we create a discriminative gait embedding space, allowing the system to verify users rather than performing complex identification. This makes it possible to enroll new users quickly with just a few examples. Testing on the challenging CASIA-B dataset shows that our system performs exceptionally well, achieving a verification accuracy of 97.51% at the optimal threshold. Overall, this framework offers a reliable and efficient solution for secure access control and continuous surveillance applications.