Through-Wall Human Trajectory Tracking Based on Multi-view Fusion Attention Mechanism

Identifying human position in through-wall scenarios is quite helpful for emergency rescue, while the multiple reflections between the wall and the radar signal substantially degrade the detection accuracy in the through-wall scenario, making it challenging to determine the human’s position. In this paper, we propose a novel approach for through-wall human trajectory tracking with radar. The group convolution mechanism is used to extract features of different views. And the soft-squeeze excitation (SSE) module is proposed to fuse multi-view features, which recalibrates channel weights by explicitly modeling the correlation between channels. With these components, the features from the bird’s eye view and front view are combined to emphasize the human’s 3D location. This method is validated on a radar-based through-wall human trajectory dataset and shows promising results.