Gait Recognition by Jointing Transformer and CNN

Gait recognition is a biometric technology based on the human walking state. Unlike other biometric technologies, gait recognition can be used for remote recognition and the human walking pattern cannot be imitated. Gait recognition has wide applications in the field of criminal investigation, security and other fields. Most of the current mainstream algorithms use Convolutional Neural Network (CNN) to extract gait features. However, CNN only captures the local image features in most cases which may not inherently capture global context or long-range dependencies. In order to solve the above problems and to extract more comprehensive and precise feature representations, we propose a novel Gait recognition algorithm jointing Transformer and CNN by introducing the attention mechanism, called GaitTC. The framework consists of three modules, including the Transformer module, CNN module and feature aggregation module. In this paper, we conduct the experiments on CASIA-B dataset. The results of the experiments show that the proposed gait recognition method achieves relatively good performance.