JRPM: Joints Relationship Pyramid Mapping for Skeleton-based Gait Recognition

One essential problem in skeleton-based gait recognition is how to extract discriminative features over all skeleton joints. However, the existing skeleton-based gait recognition models ignore the physiological relationship between joints in feature extraction, which is significant for extracting discriminative features. In this paper, the Joint Relationship Pyramid Mapping (JRPM) is proposed to increase the distinguishability of gait features. It enhances the local gait features of the human body based on the physiological relationship between the joints when human walking, that is, the physical connection and coordinated motion relationship. In addition, we propose a novel skeleton-based model Gait-J based on JRPM. Experiments show that with the help of JRPM, Gait-J has excellent performance in complex scenarios, namely, crossview and cross-walking-condition.