IEEE - Aerospace and Electronic Systems - December 2022 - 39

Fioranelli et al.
Figure 1.
Examples of spectrograms for the six activities in the dataset " Radar signatures of human activities, " namely, (a) walking, (b) sitting down
on a chair, (c) standing up from a chair, (d) picking up an object from the floor, (e) drinking a glass of water, and (f) frontal fall. Note that
the Y-axis shows the radial velocity calculated from the corresponding Doppler values accounting for the radar carrier frequency of 5.8 GHz.
A notable proposed solution involved the generation
of a point cloud from the data, by running a cell averaging
constant false alarm rate on range-time and spectrogram
plots. This is a quite creative solution for data collected at
5.8 GHz, as point cloud processing is normally applied to
data collected with mm-wave MIMO radar, e.g., 60-GHz
systems and automotive 77-GHz systems. Furthermore,
the radar had a single transmitter and receiver with no
angular measurement capabilities; hence, the resulting
point cloud had only four features containing range, Doppler,
received power, and time instant information. Nevertheless,
this representation combined with a network
designed to work on point clouds-in this case, a simple
PointNet-provided very promising results. Part of the
processing and network development for this approach
was performed within the Seminar series of Computer
Vision by Deep Learning offered by TU Delft.2
The highest performances in terms of F1 score on
the unlabeled test dataset were achieved by two groups
of students, group #5 (Jurgen Wervers and Bob van
Nifterik) who achieved 79% and group #7 (Mujtaba
Hassan, Yanwen Chen, and Victor Oliveira) who
achieved 88%. The reports and algorithm code generated
by these two groups has been publicly shared.3 A
typical example of reported misclassification was the
confusion between the " picking up " and " drinking
water " activities, as in both cases, the participants had
2https://github.com/SimmyZhu/CS4245_Project
3https://github.com/SimmyZhu/TUD-EE4675-Project-2022
DECEMBER 2022
to extend their body and arm to take an object and
then move it closer to their bodies, ending up in very
similar micro-Doppler signatures.
In conclusion, while this student activity was performed
at a small scale and with educational rather than
research aims, we are happy to report it to the AESS
readership, as we believe that such initiatives with open
datasets sharing and classification algorithm benchmarking
are beneficial for the wider radar research
community.
REFERENCES
[1] D. Gusland, J. M. Christiansen, B. Torvik, F. Fioranelli,
S. Z. Gurbuz, and M. Ritchie, " Open radar initiative: Large
scale dataset for benchmarking of micro-Doppler recognition
algorithms, " in Proc. IEEERadar Conf., 2021, pp. 1-6,
doi: 10.1109/RadarConf2147009.2021.9455239.
[2] F. Fioranelli, S. A. Shah, H. Li, A. Shrestha, S. Yang, and
J. Le Kernec, " Radar signatures of human activities, " in
Proc. Asian Conf. Innov. Technol., 2019, pp. 1-4,
doi: 10.5525/gla.researchdata.848.
[3] F. Fioranelli, S. Aziz Shah, H. Li, A. Shrestha, S. Yang,
and J. Le Kernec, " Radar sensing for healthcare, "
Electron. Lett., vol. 37, no. 9, pp. 1022-1024, 2019,
doi: 10.1049/el.2019.2378.
[4] M. Ritchie, R. Capraru, and F. Fioranelli, " Dop-NET: A
micro-Doppler radar data challenge, " Electron. Lett.,
vol. 56, pp. 568-570, 2020, doi: 10.1049/el.2019.4153.
IEEE A&E SYSTEMS MAGAZINE
39

http://dx.doi.org/10.1109/RadarConf2147009.2021.9455239 http://dx.doi.org/10.5525/gla.researchdata.848 http://dx.doi.org/10.1049/el.2019.2378 https://github.com/SimmyZhu/CS4245_Project http://dx.doi.org/10.1049/el.2019.4153 https://github.com/SimmyZhu/TUD-EE4675-Project-2022

IEEE - Aerospace and Electronic Systems - December 2022

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - December 2022