Adaptive Kinematic Masking for Urban Human Activity Recognition Using UWB Radar: A Sim2Real Approach

Abstract

Ultra-wideband (UWB) radar is promising for human activity recognition (HAR), but its coarse micro-Doppler resolution and the scarcity of labeled data hinder practical application. To address these challenges, we propose an adaptive kinematic masking method under a simulation-to-reality (Sim2Real) framework. First, an adaptive thresholding approach is employed to generate precise kinematic masks from dual-channel micro-Doppler heatmaps. Subsequently, our method utilizes a lightweight mask decoder as a regularization branch to enforce motion-aware feature learning. The effectiveness of the proposed method is validated through a measured dataset collected for the UWB radar HAR task, demonstrating enhanced multi-scene generalization without additional inference overhead.