End-to-End Human Pose Reconstruction from Radar Signals Based on Deep Residual Networks

The reconstruction of human pose using radar primarily focuses on effectively estimating the position and pose of a target based on radar echo signals. Compared to traditional optical methods, radar signals can operate under limited lighting conditions, in the presence of occlusions, or beyond the line of sight. This paper proposes an end-to-end human pose reconstruction method based on a deep residual network. This method bypasses traditional imaging and detection algorithms, directly extracting amplitude and phase features from radar echo signals, and outputs the target's azimuth, distance, and pose information. Experimental results verify the effectiveness of the algorithm.