Radar Human Motion Classification Using Multi-Antenna System Patrick A Schooley and Syed A. Hamza School of Engineering, Widener University, Chester, PA 19013, USA ABSTRACT This paper considers human activity classification for an indoor radar system. Human motions generate non- stationary radar returns which represent Doppler and micro-Doppler signals. The time-frequency (TF) analysis of micro-Doppler signals can discern subtle variations on the motion by precisely revealing velocity components of various moving body parts. We consider radar for activity monitoring using TF-based machine learning approach exploiting both temporal and spatial degrees of freedom. The proposed approach captures different human motion representations more vividly in joint-variable data domains achieved through beamforming at the receiver. The radar data is collected using real time measurements at 77 GHz using four receive antennas, and subsequently micro-Doppler signatures are analyzed through machine learning algorithm for classifications of human walking motions. We present the performance of the proposed multi antenna approach in separating and classifying two closely walking persons moving in opposite directions. 1. INTRODUCTION Human motion recognition (HMR) finds important applications in a large variety of scenarios ranging from ges- ture recognition for smart homes, detecting events of interest for automatic surveillance, behavioral analysis, Gait abnormality recognitions, health monitoring in care facilities and rehabilitation services to enable independent living for elderly. 1–6 Contactless sensing of human motions has gained traction because of the obvious benefits of being non obtrusive. It does not require any user intervention and as such the users are not required to wear specific devices to be tracked via smart phone applications. 7–9 Radar systems are at the forefront of remote sensing technologies as they provide robust non contact monitoring that is not affected by lighting conditions. Additionally, active RF (radio frequency) sensing provides 4D imaging capabilities by explicitly measuring the scatterer velocity in addition to range and 2D angular localization. This is unlike other remote sensing sensors of human motions such as visual-based systems that require additional pre-processing and filtering operations to accurately discern small movements. 10, 11 Also radar images are privacy preserving as high resolution imaging radar renders silhouette type portrait revealing little identifiable information as opposed to camera-based systems. We investigate a human activity monitoring system to concurrently monitor movements of multiple persons. This could facilitate to separately record the activities of multiple persons in detail. For example, in health care facility, the task of care givers can be eased by attending to the needs of several care receivers at the same time. Radar is a proven technology for target detection, localization and tracking. Imaging radars are getting attention recently because of their added capability of classifying different targets. Radar returns classification could be performed after localizing the target in range, azimuth and/or Doppler. In this paper, we consider the radar human motion classification by attempting to localize the motions to a given azimuth directions. The proposed beamforming approach can reduce the system cost and alleviate the need of using multiple radars as proposed in. 10, 12–14 Azimuth filtering is achieved by applying beamforming to the receiver array. Specifically, we consider the task of classifying two persons walking closely in opposite directions at different azimuth angles. We aim to correctly pair the direction of motion to the corresponding azimuth angle. This is achieved by jointly processing the two spectrograms obtained in the directions of both motions. In this case, the received data is filtered using beamforming with two separate sets of beamformer E-mails: {paschooley, shamza}@widener.edu arXiv:2104.00217v1 [eess.SP] 1 Apr 2021