MsfGait: A Multimodal Gait Recognition Method

In gait recognition, for its susceptibility to clothing occlusion and carrying objects, this paper proposes a multimodal gait recognition method MsfGait based on multiscale temporal feature fusion, which extracts the frame-level features of the contour map and the skeleton map through two independent feature extraction modules, this paper designs a spatial feature extraction module, which extracts the spatial features of the contour and the skeleton, respectively, and designs a multiscale temporal feature extraction module, which introduces an attention mechanism to extract short-term and long-term temporal features of multimodal feature maps using independent branches; experimental results on CASIA-B and OUMVLP datasets show the effectiveness of the method proposed in this paper, and there is a large increase in the robustness and recognition success rate of the model in the presence of occlusion.