ProbRadarM3F: mmWave Radar based Human Skeletal Pose Estimation with Probability Map Guided Multi-Format Feature Fusion

• DOI: 10.48550/arxiv.2405.05164
• Publication Date: 2024-05-08
• Authors: Bing Zhu; Zixin He; Weiyi Xiong; Guanhua Ding; Jianan Liu; Tao Huang; Wei Chen; Wei Xiang
• Keywords: Feature (linguistics); Sensor Fusion

Millimetre wave (mmWave) radar is a non-intrusive privacy and relatively convenient inexpensive device, which has been demonstrated to be applicable in place of RGB cameras human indoor pose estimation tasks. However, mmWave relies on the collection reflected signals from target, containing information difficult fully applied. This long-standing hindrance improvement accuracy. To address this major challenge, paper introduces probability map guided multi-format feature fusion model, ProbRadarM3F. novel extraction framework using traditional FFT method parallel with based positional encoding method. ProbRadarM3F fuses heatmap features features, then effectively achieves 14 keypoints body. Experimental evaluation HuPR dataset proves effectiveness model proposed paper, outperforming other methods experimented an AP 69.9 %. The emphasis our study focusing position that not exploited before singal. provides direction investigate potential non-redundant rader.