ConvNeXt network with transfer learning for cumulative foot pressure images recognition

Abstract

Humans can be differentiated by their walking style using cumulative foot pressure images, which are 2-D cumulative ground reaction forces during each gait cycle. In this paper, we propose a new model for cumulative foot pressure images in which we apply transfer learning to our ConvNeXt model after adjusting it based on our usage to acquire discriminative features that can identify humans using only cumulative foot pressure images. Our experiment was conducted on CASIA Dataset D using only cumulative foot pressure images of eighty-eight persons while neglecting the gait pose images captured by the camera. The evaluation indicates that our proposed method is effective, where we achieved an accuracy of 99.894%, a false acceptance rate of 0.002%, and a false rejection rate of 0.039%, which shows a significant improvement over the previous work, concluding that the cumulative foot pressure images have the potential to be used as a biometric identifier.