Radar Network for Gait Monitoring: Technology and Validation

TL;DR

It is shown that analyzing the feet velocity increases the reliability of the temporal parameters, especially with aged or motorically impaired subjects, and is significant for the implementation of radar networks in clinical and domestic environments.

Abstract

In recent years, radar-based devices have emerged as an alternative approach for gait monitoring. However, the radar configuration and the algorithms used to extract the gait parameters often differ between contributions, lacking a systematic evaluation of the most appropriate setup. Additionally, radar-based studies often exclude motorically impaired subjects, leaving it unclear whether the existing algorithms are applicable to such populations. In this paper, a radar network is developed and validated by monitoring the gait of five healthy individuals and three patients with Parkinson's disease. Six configurations and four algorithms were compared using Vicon as ground-truth to determine the most appropriate solution for gait monitoring. The best results were obtained using only three nodes: two oriented towards the feet and one towards the torso. The most accurate stride velocity and distance in the state of the art were obtained with this configuration. Moreover, we show that analyzing the feet velocity increases the reliability of the temporal parameters, especially with aged or motorically impaired subjects. The contribution is significant for the implementation of radar networks in clinical and domestic environments, as it addresses critical aspects concerning the radar network configuration and algorithms.

Figures (14)

• Figure 1: 24-GHz radar node. (a) RF and Control PCBs with the external elements that interact with the radar. (b) Schematic of the node. (c) Antenna characterization (left) and radar EIRP (right).
• Figure 2: Signal model of LFMCW radars, where $f_0$ is the start frequency, $B$, the bandwidth, $T_c$, the chirp time, $R$, the range of the target and $c$, the speed of light.
• Figure 3: Algorithm implemented to extract the range-time and Doppler-time matrices.
• Figure 4: Radar network implementation. (a) Illustration of the radar network with a subject performing the TUG test. (b) Photograph of the experimental setup highlighting the nodes and the Vicon cameras used as ground-truth.
• Figure 5: Vicon wearable markers have minimal impact on the SNR of the Doppler-time matrix. The signal power distribution with and without them is shown in the bottom figure, displaying a similar SNR in both cases.