RF-PSignal: The Smart Home Gesture Recognition System Based on Channel-Wise Topology and Self-Adaptive Attention

With the rapid development of the Internet of Things, smart home technology is bringing unprecedented convenience to the lives of the elderly. Particularly, applications for controlling smart homes based on Radio Frequency Identification (RFID) technology are opening up broad prospects for human-computer interaction. However, traditional RFID-based gesture recognition technologies are constrained by antenna beamwidth, posing certain challenges for elderly users in precisely controlling smart homes. To address this, this paper proposes a contactless RF-PSignal system based on Commercial Off-The-Shelf RFID devices. This system achieves high accuracy and lightweight gesture recognition for smart homes across wide areas. First, we fuse the signal phase and signal strength and propose a time-series-to-image algorithm to generate denoised images. Second, the system proposes three feature extraction modules designed based on the concept of channel topology, effectively improving recognition accuracy while reducing model complexity. Finally, we innovatively propose an adaptive attention module that dynamically assigns attention weights to enhance system performance. Experimental results show that the system achieves an average gesture recognition accuracy of 95.35% in wide-area scenarios. Furthermore, the system attains an average recognition accuracy of 97.53% in cross-environment scenarios and 96.22% in cross-user scenarios. Unlike most existing studies that rely on time-series signals for feature extraction and pattern recognition, the system achieves outstanding performance and computational efficiency through a time-series-to-image transformation algorithm and lightweight model design. Experimental results across various complex scenarios demonstrate the system’s generalizability and high robustness.