A Dual-View UWB Arm Motion Recognition System With Morphological Filtering and Spatiotemporal Fusion

Ultra Wideband (UWB) signals offer high spatio-temporal resolution, penetrability, and low cost, which facilitates accurate characterization of limb features through micro-Doppler (mD) analysis, even with random body movements. Due to challenges introduced by arm motions, which may be perpendicular to the radar, we introduce a dual radar arm motion recognition system with light-weighted feature extraction and appropriate data fusion. This architecture can enhance the motion characterization by integrating complementary perspectives from the front and side radar nodes. Through proper morphological filtering and motion feature extraction, we adopt the spatio-temporal feature fusion and Doppler signatures to enhance the diverse arm motion recognition via attention-guided weighting. Meanwhile, comparisons against spatial, temporal, and graph-based learning models validate the proposed method. According to practical experiments, this system can achieve 92.81% accuracy for 10 typical arm motions with limited data sets. Compared to the single radar system, over 10% accuracy improvements can be achieved by introducing an additional radar. Furthermore, it shows that the system can significantly benefit from the proposed morphological filtering and motion feature extraction, with about 30% accuracy enhancement. It validates the effectiveness of dual-view sensing for human motion recognition and also provides valuable insights for future ubiquitous wireless sensing systems.