Exploring transformers for behavioural biometrics: A case study in gait recognition

• DOI: 10.1016/j.patcog.2023.109798
• Publication Date: 2023-07-04T00:59:47Z
• Authors: Paula Delgado-Santos; Ruben Tolosana; Richard Guest; Farzin Deravi; Ruben Vera-Rodriguez
• Keywords: Robustness

Biometrics on mobile devices has attracted a lot of attention in recent years as it is considered user-friendly authentication method. This interest also been motivated by the success Deep Learning (DL). Architectures based Convolutional Neural Networks (CNNs) and Recurrent (RNNs) have established convenience for task, improving performance robustness comparison to traditional machine learning techniques. However, some aspects must still be revisited improved. To best our knowledge, this first article that explores proposes novel gait biometric recognition systems Transformers, which currently obtain state-of-the-art many applications. Several architectures (Vanilla, Informer, Autoformer, Block-Recurrent Transformer, THAT) are experimental framework. In addition, new Transformer configurations proposed further increase performance. Experiments carried out using two popular public databases: whuGAIT OU-ISIR. The results achieved prove high ability outperforming CNN RNN architectures.