A Study of on-Body RF Characteristics Based Human Body Motion Detection

Detecting human body motion is a promising application to monitor the health status of the elderly or patient. With the rapid advancement of the fifth generation of cellular network technology (5G), an increasing number of radio frequency (RF) devices are deployed on the human body. These devices have great potential to continuously and remotely detect human motions. In this paper, we propose a thorough study for developing effective on-body RF motion detection systems from three perspectives, including effects of signal frequencies, effects of on-body sensor locations, and effects of classification algorithms. First, based upon handcrafted on-body RF characteristics, the detection accuracies are studied at 0.4GHz, 2.4GHz, and 5GHz, when the RF transceiver is mounted on the human hand, foot, and head. A minimal redundancy maximal relevance (mRMR) based stepwise detection is implemented to cross-validate the study. It also proves that a subset of candidate features can provide an acceptable detection accuracy. Moreover, to improve the detection performance, an attention-based bidirectional long-short term memory (BiLSTM) is proposed for on-body RF motion detection. We distinguish the typical human body motions, including standing, walking, and falling, with an average accuracy of 99.04%. The performance of the proposed attention-based BiLSTM has been analyzed with visualized training histories and attention weights.