User Identification Under the Collaborative Auto-Calibration of Multi-mmWave Radars

Person tracking and recognition systems play a key role in the field of smart device-free sensing, as they can provide personalized services for specific individuals. Cameras are commonly used sensors to implement these systems, but their privacy issues have raised concerns. Millimeter-wave (mmWave) radar has the advantages of low cost and privacy protection. The focus of this article is to use low-cost mmWave radar to achieve user identity (ID) recognition. To address the challenge posed by the sparsity and limited coverage of single radar point clouds in user feature extraction, we utilize the collaboration of multiple mmWave radars to achieve high-precision user ID recognition with a broader coverage range. We propose a fast auto-calibration method that efficiently transforms the point clouds captured by different radar devices into a unified coordinate system. A novel deep neural network, dubbed MRINet, is introduced to extract spatiotemporal features from radar point cloud sequences, thereby enabling high-precision user identification. Evaluation results show that the calibration errors of the device’s position and heading for the proposed method are 0.1 m and 2°, respectively, and the user ID recognition accuracy is 97.27% in multiperson recognition tasks. Compared to the two existing models, one classical and one state-of-the-art, our model achieves the best performance.