Skeleton-based human activity recognition has become one of the research hotspots in the field of pattern recognition. However, existing methods have limitations in achieving fine-grained capture of human activity spatiotemporal features, such as (1) being unable to distinguish which neighboring nodes are important to the current node; (2) Unable to capture local temporal features. In this article, we utilize RFID, which has the advantages of privacy protection, identifiability, and battery free maintenance, to perceive human activities. We design a body RFID skeleton graph to express human activities and propose a body RFID skeleton spatiotemporal graph convolutional network (BRS-GCN) to capture the spatiotemporal features of the body RFID skeleton graph at a fine-grained level, thereby achieving human activity recognition. The recognition performance advantage of BRS-GCN compared to existing HARs has been verified through experiments.