IEEE Consumer Electronics Magazine - January/February 2024 - 77

fast, accurate, reliable, and privacy-aware detection.
However, further experimentation and validation
are needed to fully realize the potential of
the proposed framework. In this section, we provide
insights into expected outcomes.
The study in Jiang et al.6 evaluates multiple
versions of YOLO for multiobject detection and
finds YOLOv5-s to have the best overall performance
in termsof mAP, precision, speed, and storage
when tested on the FLIR ADAS thermal
dataset, with values of 0.803, 0.638, 41 FPS,
and 14 MB, respectively. Li et al.14
supported
the improved performance of the YOLOv5s_Ghost_CAmodel
for facial expression detection,
with a boost in mAP0:5 from 98.4 to 98.8 and maintained
mAP0:5:0:95. This results in reduced weights
(15.4 to 8 MB), parameters (7.02 to 3.70 M), and
computation cost (15.8 to 8.1 GFLOPs), with
improved inference time (115 to 123 FPS). EdgeYOLO11
improved FPS to 26.6 from 4.9 onYOLOv4,
with COCO2017 dataset, powered on Jetson Xavier.
With the KITTI dataset, EdgeYOLO improved
FPS to 40.6 from 5.2 on YOLOv4, powered on Jetson
Xavier. Jetson Nano, with its lower memory
and computation power, showed lower FPS than
Jetson Xavier, but still demonstrated improved
speed between YOLOv4 and EdgeYOLO. Our
framework utilizes backbone of the YOLOv5s_Ghost_CA
model14 with modified neck from
EdgeYOLO11 combined with online learning. The
proposed model combining lightweight YOLOv5-s
with real-time stateful learning has the potential
to optimize weights, parameters, and computation
costs while maintaining a high mAP score.
Moreover, it can improve latency and reduce
energy consumption, based on the promising
results of these approaches.
CONCLUSION
This article proposes a real-time online physical
threat detection model using edge AI computing
and online learning. The experimental nature
of real-time machine learning, manual data generation,
and multiple parameters pose a challenge
in achieving optimal results. Repeated
experimentation using optimized models and
advanced edge hardware could soon realize this
vision, enhancing speed, accuracy, reliability,
and privacy in computing technology.
January/February 2024
ACKNOWLEDGMENTS
This work was supported by the DoD Center
of Excellence in AI and Machine Learning (CoEAIML)
at Howard University under Contract
Number W911NF-20-2-0277 with the U.S. Army
Research Laboratory and in part by the US NSF
under Grant CNS/SaTC 2039583.
&REFERENCES
1. IANS, " At 12 billion, IoT connections to surpass nonIoT
devices in 2020, " The Times of India, 2020.
Accessed: Oct. 30, 2022. [Online]. Available: https://
telecom.economictimes.indiatimes.com/news/at-12billion-iot-connections-to-surpass-non-iot-devices-in2020/79318722
2.
W. Z. Khan, E. Ahmed, S. Hakak, I. Yaqoob, and A.
Ahmed, " Edge computing: A survey, " Future Gener.
Comput. Syst., vol. 97, pp. 219-235, 2019.
3. S. C. Hoi, D. Sahoo, J. Lu, and P. Zhao, " Online learning:
Acomprehensive survey, " Neurocomputing, vol. 459,
pp. 249-289, 2021, doi: 10.1016/j.neucom.2021.04.112.
4. J. Redmon, S. Divvala, R. Girshick, and A. Farhadi,
" You only look once: Unified, real-time object
detection, " in Proc. IEEEConf. Comput. Vis. Pattern
Recognit., 2015, pp. 779-788.
5. Dhiraj and D. K. Jain, " An evaluation of deep learning
based object detection strategies for threat object
detection in baggage security imagery, " Pattern
Recognit. Lett., vol. 120, pp. 112-119, 2019,
doi: 10.1016/j.patrec.2019.01.014.
6. C. Jianget al., " Object detection from UAV thermal
infrared images and videos using YOLO models, " Int.
J. Appl. Earth Observ. Geoinf., vol. 112, 2022,
Art. no. 102912, doi: 10.1016/j.jag.2022.102912.
7. A.Warsi, M. Abdullah, M. N. Husen, M. Yahya, S. Khan,
and N. Jawaid, " Gun detection systemusingYOLOv3, " in
Proc. IEEEInt. Conf. SmartInstrum., Meas. Appl., 2019,
pp.1-4,doi: 10.1109/ICSIMA47653.2019.9057329.
8. P. K. Donta, S.N. Srirama, T. Amgoth, and C. S. R.
Annavarapu, " Survey on recent advances in IoT
application layer protocols and machine learning
scope for research directions, " Digit. Commun. Netw.,
vol. 8, no. 5, pp. 727-744, 2022.
9. B. Dowlen, " Scylla AI-powered gun detection system
integrated with intrepid networks response platform, "
Aug. 5, 2021. Accessed: Feb. 5, 2023. [Online].
Available: https://www.einnews.com/pr_news/
547342913/scylla-ai-powered-gun-detection-systemintegrated-with-intrepid-networks-response-platform
77

https://telecom.economictimes.indiatimes.com/news/at-12-billion-iot-connections-to-surpass-non-iot-devices-in-2020/79318722 https://telecom.economictimes.indiatimes.com/news/at-12-billion-iot-connections-to-surpass-non-iot-devices-in-2020/79318722 https://telecom.economictimes.indiatimes.com/news/at-12-billion-iot-connections-to-surpass-non-iot-devices-in-2020/79318722 https://telecom.economictimes.indiatimes.com/news/at-12-billion-iot-connections-to-surpass-non-iot-devices-in-2020/79318722 https://doi.org/10.1016/j.neucom.2021.04.112 https://doi.org/10.1016/j.patrec.2019.01.014 https://doi.org/10.1016/j.jag.2022.102912 https://doi.org/10.1109/ICSIMA47653.2019.9057329 https://www.einnews.com/pr_news/547342913/scylla-ai-powered-gun-detection-system-integrated-with-intrepid-networks-response-platform https://www.einnews.com/pr_news/547342913/scylla-ai-powered-gun-detection-system-integrated-with-intrepid-networks-response-platform https://www.einnews.com/pr_news/547342913/scylla-ai-powered-gun-detection-system-integrated-with-intrepid-networks-response-platform

IEEE Consumer Electronics Magazine - January/February 2024

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - January/February 2024