1/5

SDN-based Internet of Video Things Platform

Enabling Real-Time Edge/Cloud Video Analytics
Orest Kochan Mykola Beshley Halyna Beshley
School of Computer Science Department of telecommunications Department of telecommunications
2023 17th International Conference on the Experience of Designing and Application of CAD Systems (CADSM) | 979-8-3503-1085-6/23/$31.00 ©2023 IEEE | DOI: 10.1109/CADSM58174.2023.10076539

Hubei University of Technology Lviv Polytechnic National University Lviv Polytechnic National University
Wuhan, China Lviv, 12 Bandery Str., Ukraine Lviv, 12 Bandery Str., Ukraine
Department of Information-Measuring Department of Information Systems, Department of Information Systems,
Technologies Faculty of Management, Faculty of Management,
Lviv Polytechnic National University Comenius University in Bratislava, Comenius University in Bratislava,
Lviv, 12 Bandery Str., Ukraine 82005 Bratislava, Slovakia 82005 Bratislava, Slovakia
E-mail: [email protected] E-mail: [email protected] E-mail: [email protected]

Yuriy Shkoropad Iryna Ivanochko Nadiia Seliuchenko

Department of telecommunications Department of Management and Department of Business Economics and
Lviv Polytechnic National University International Business Investment Lviv Polytechnic National
Lviv, 12 Bandery Str., Ukraine Lviv Polytechnic National University University
E-mail: [email protected] Lviv, 12 Bandery Str., Ukraine Lviv, 12 Bandery Str., Ukraine
E-mail: [email protected] E-mail: [email protected]

Abstract—The trend of recent years is the continuous traffic, determine the effectiveness of the created urbanism
development of the Internet of Things (IoT). Among such things, using statistical data, and automatically record offenses;
a significant share is occupied by visual sensors and video search for missing people; automatic notifications and quick
cameras that generate large amounts of data. In turn, the need responses in case of danger, for example, military or terrorist
to attract significant storage resources, transmission invasion, vandalism, fire, explosion, earthquake, etc;
throughput, and processing power is an inevitable solution for monitoring the observance of social distancing and mandatory
real-time video analytics. Thus, the combination of smart wearing of masks (during a pandemic). Thus, using
cameras with the computing paradigm of Cloud/Edge and IoT surveillance cameras in smart cities will significantly improve
architectures form the next generation of video surveillance
the safety, discipline, and comfort of its residents and
systems, called the "Internet of Video Things" (IoVT In this
paper, a new IoVT platform is developed that, in addition to
significantly save the city budget [7-9]. The installation of
harmoniously combining Edge/Cloud computing, uses SDN to cameras does not require the immediate installation of all the
overcome challenges such as flexible management, control, and necessary cameras in all places and the provision of servers
maintenance of IoVT devices. In particular, within the proposed with programs to solve all analytical problems. The city can
IoVT platform, an algorithm for the dynamic selection of Edge be provided with cameras in stages, just like analytical tasks.
or Cloud computing is implemented using an SDN controller to This task also requires special attention to the security of such
provide effective video analytics in real-time. This algorithm a system. After all, an attacker can fabricate evidence, disable
considers such parameters as the priority of computational cameras, or gain access to classified information. An equally
tasks, the number of video streams, and the image quality with important attribute for the system will be the ability to
the ability to adapt to a specific application by software prioritize tasks. For example, when a serious crime has
configuration of the IoVT platform. We also demonstrate the occurred and it is necessary to track the criminal, computing
effectiveness of the proposed solutions on real equipment and servers should put aside tasks related to collecting statistics.
discuss several promising areas of application of the developed With such requirements for the system, it is possible to define
platform. such non-functional requirements for the system as scalability,
security, modularity, prioritization, and fast data processing
Keywords—Internet of Video Things, Software-defined [10-15].
Networking, Cloud Computing, Edge Computing, Quality of
Service. The best way to ensure that the system can easily scale is
to use the computing resources of cloud services, which make
I. INTRODUCTION it possible to scale any number of computing servers in a very
The development of video analytics technology requires short period of time [16]. However, some tasks require a quick
the transformation of the modern Internet of Things (IoT) [1] response and may also require better data protection [17]. For
to create new opportunities for flexible and high-quality use this purpose, the system can get these advantages by using
in everyday life. Thus, the combination of various video edge computing. Since the data is physically located nearer to
sensors [1] and cameras in a single IoT environment creates a the cameras and does not require Internet access, video
new networking technology trend called the Internet of Video processing and analysis will be much faster and more secure.
Things (IoVT) [2-4]. Surveillance plays a very important role An effective solution is the hybrid use of Edge/Cloud
in modern smart cities [5], which in turn drives the continuous computing, which in turn requires IoVT platforms to be able
development of IoVT platforms. There are many needs for the to continuously control and monitor the status of resources to
state and private enterprises to use it [6]: to ensure security in implement an adaptive selection of the computing domain and
the city by recognizing suspicious people and guarding in real- prioritize tasks [18, 19]. Software-defined networking (SDN)
time, and to analyze video after an incident; to control allows software control of the network, which means it can be
car

979-8-3503-1085-6/23/$31.00 ©2023 IEEE

Authorized licensed use limited to: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on April 15,2023 at 19:05:31 UTC from IEEE Xplore. Restrictions apply.
 1/6

easily reconfigured and automated [20]. The convergence of performance and more flexible transmission capabilities. The
Edge/Cloud computing with SDN allows IoVT platforms to Edge computing system is based on the single-board
be both flexible and cognitive, transforming them into computers Raspberry Pi [23].
autonomous systems [21]. In such platforms, higher-priority
video analytics tasks will be performed on edge servers, and The advantage of the proposed IoVT platform compared
lower-priority tasks on cloud servers. to the authors' work [24] is the ability to schedule some video
analytics tasks and perform them gradually. To do this, the
II. SDN-BASED INTERNET OF VIDEO THINGS PLATFORM platform offers a message broker that will receive new
WITH EDGE/CLOUD СOLLABORATIVE СOMPUTING analytical tasks, and cloud servers will perform the next tasks
in the queue as they are completed. Since the message broker
This paper proposes a new IoVT platform that leverages stores notifications about specific tasks, the videos themselves
SDN to harmonize Edge/Cloud computing and address issues will be stored in file storage (bucket storage). This will also
such as flexible management, control, and support of IoVT allow the administrator to manually analyze the required video
devices [22]. Taking into account a number of needs that may segment. Initially, videos will be stored in temporary local file
arise in practice, the architecture of this system was developed storage, from where they will be automatically sent to the
in such a way as to scale this system without additional cloud file storage. Edge servers will use local storage, and
problems in the future, namely, the ability to connect multiple cloud servers will use cloud storage. The results of all
cameras and users without degrading QoS. To this end, the calculations from the edge and cloud servers will be stored in
developed network prototype was used to implement intuitive the database. To manage all this, the user will have convenient
inter-server edge/cloud load balancing in the face of QoS access through the mobile and web application, as well as
degradation and real-time video image analysis. Given that receive Push notifications for quick response if necessary.
this system is designed for a mission-critical service, it also Tasks will also be created using the app. The new task will be
implements a reverse proxy that optimizes traffic, encrypts added to the database, from where the free server will take
packets, balances the load, and protects against distributed over the task. The structural and functional diagram of the
denial-of-service attacks. The key element for the scalability proposed SDN-based IoVT platform for real-time video
of the proposed IoVT platform is the OpenFlow controller and analytics is shown in Fig. 1.
the Zodiac FX hardware SDN switch, which, provides better

Fig. 1. Proposed 5G network architecture

The users will have the possibility to see the video in real- III. ALGORITHM ENABLING REAL-TIME EDGE/CLOUD
time with an acceptable delay. The received video stream from
COMPUTING VIDEO ANALYTICS FOR SDN-BASED IOVT
the cameras is sent to the SDN switch, from where it will
PLATFORM
either go directly to the cloud server or to the edge server. In
the case of the edge server, the video is analyzed, and the We propose an algorithm in which the SDN, edge server
results of the calculations are stored in the cloud database. and cloud server cooperate with each other to manage the
video stream to support the real-time analytics process,
The SDN controller determines whether the task should be including in conditions of limited network throughput. Our
performed on the cloud or the edge server. The process of key idea is that the SDN controller distributes data flows in
determining the controller is realized by analyzing the degree SDN between the Edge and Cloud computing domains based
of server utilization (CPU, RAM). The servers transmit data on task importance scores that indicate the priority of the data
on the load level to the SDN controller in separate packets from a video analytics perspective. The generalized algorithm
with a certain monitoring step. of the SDN controller is shown in Fig. 2.

Authorized licensed use limited to: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on April 15,2023 at 19:05:31 UTC from IEEE Xplore. Restrictions apply.
 1/7

error. Therefore, for better accuracy, make sure that the clocks
on all modules are synchronized with the Network Time
Protocol (NTP) [25].

Fig. 3. SDN based IoVT platform prototype

Since end users mainly use the system through the

web application, we investigated video viewing delays in the
case when the web application is deployed on a local
network closer to the cameras themselves for private video
surveillance [26]. And also when the web application is
deployed on the global network (Internet) to monitor video
from any access point. The web application is deployed on a
global network on free hosting on Google's Firebase
[27-29]. Fig. 4 shows a graph of delays in a web
application.

Fig. 2. SDN controller operation algorithm enabling real-time edge/

cloud computing video analytics

For a new task, the controller checks whether the

edge servers' utilization rate is below a certain level; if so,
the task will proceed immediately. If not, the edge
server will be instructed to pause the task if it is a lower
priority and transfer it to the cloud server. If the server has
enough resources to perform the high-priority task, it will
immediately start the task. Some tasks have quality
requirements for the minimum video resolution at which the
system can recognize the objects you are looking for in the
video with sufficient quality. In such cases, when the
Internet speed deteriorates significantly, or when all
resources are occupied, and it is necessary to add another
task, you can force the video quality to be reduced, Fig. 4. SDN based IoVT platform prototype
provided that the work being performed allows it.
The delay shows the time it takes for a packet to
Also, some tasks can be performed by different AI travel from the video stream server (edge 1) to the web
models, which may differ in quality, reliability of object application. As a result, when 10 clients are connected, a
detection, or other tasks. However, inferior models can slight overall delay is created (in the range of 10-50 ms),
perform these tasks much faster and with less load on which is also true for both the case of connecting via a local
the server's CPU. Sometimes the quality of weaker models web application and via the Internet. Unlike the Internet web
is acceptable for the task. When the servers reach a certain application, the local web application does not need to go
level of load, which leads to a violation of the condition through intermediate connections, and therefore has a
for providing real-time video analytics, optimization is shorter delay; according to the graph, we can roughly
performed, for example, by reducing the video resolution or estimate this delay in the range of 150-250 ms. It is also
switching to a simpler model for all tasks performed at that worth testing for a web application how many requests it
moment in time. can process simultaneously. To do this, we used the ab
utility is an Apache HTTP server command to test server
IV. BENCHMARKING IOVT PLATFORM loads by creating several parallel requests. The standard
The prototype of the proposed IoVT platform for library settings were used, 1000 requests, 100 requests
research and benchmarking is shown in Fig. 3. To in parallel. The results are shown in Fig. 5. So, the average
evaluate the performance of the prototype of this system, time to establish a connection was 193 milliseconds,
the level of video stream delays can be measured. It should request processing was 71 milliseconds, waiting was 50
be noted that the measurements of delays between milliseconds, and the total time was 264 milliseconds. The
different elements of the system are made by the internal data shown is for making an https request to a web
clock, which may have an application.

Authorized licensed use limited to: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on April 15,2023 at 19:05:31 UTC from IEEE Xplore. Restrictions apply.
 1/8

At the same time, the load on the system running the video efficiency of the proposed approach to the adaptive selection of
service and the video stream was also compared in parallel. a video service server in terms of QoS is proved.
Using mpstat, a command of Unix-like operating systems for
obtaining statistics related to the processor, we analyzed the
load of the system (in this case, Raspberry Pi edge server 2).
Fig.6 shows the result of this function. As you can see, the
number of active connections is within acceptable limits and
does not increase the system load.

а)

b)
Fig. 7. Estimating video surveillance delay under critical load of edge
server 1 (a) and estimating video surveillance delay under adaptive
selection of less loaded edge server 2 (b)
Fig. 5. The result of load testing with the ab program
We also tested the prototype IoVT platform to evaluate
the delay of video analytics using edge and cloud servers.
For this purpose, we connected another camera module to
the IoT gateway. The delays of the prototype of this system
are shown in Fig. 8 and Fig. 9.

Fig. 8. Video analytics delays with edge servers

Fig. 6. Edge video server 2 load statistics

The next stage of testing was to verify the

implementation of the idea of choosing the optimal service
time for the edge video server. In particular, on edge video
server 1, its load was forced to reach the critical level of
CPU operation. Fig.7a shows the delay has risen sharply to
unacceptable levels (128489 ms), and the frames per
second value is very low, which makes it impossible to
use this system. As a result, real-time video viewing
was impossible. Thus, using the proposed algorithm, the
client's service is automatically switched to edge video
server 2, which has enough resources to view the video in Fig. 9. Video analytics delays with cloud servers
real-time. In this case, the delay fluctuates slightly more
than a second (1124 ms), but it is quite acceptable for
use (Fig. 7b). Thus, the

Authorized licensed use limited to: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on April 15,2023 at 19:05:31 UTC from IEEE Xplore. Restrictions apply.
 1/9

The graph shows when using edge servers/cloud servers, Computer Engineering (TCSET), Lviv-Slavske, Ukraine, 2018, pp.
computing near the camera gives better results, but due to the 1118-1122.
limitation of computing resources, the number of such [11] H. Xu, et al. “A Combination Strategy of Feature Selection Based on
an Integrated Optimization Algorithm and Weighted K-Nearest
connected cameras with the tasks performed does not scale Neighbor to Improve the Performance of Network Intrusion
well, so it is best to use a hybrid solution with cloud services Detection,” Electronics, vol. 9, no. 8, p. 1206, Jul. 2020
based on the video analytics logic of the proposed algorithm. [12] M. Klymash, R. Savchuk, P. Pozdnyakov and M. Beshley, "The
researching and modeling of structures of mobile networks for
V. CONCLUSION providing of multiservice radio access," Proceedings of International
Conference on Modern Problem of Radio Engineering,
It is established that edge computing is widely used in the Telecommunications and Computer Science, Lviv, Ukraine, 2012, pp.
analysis of streaming video data in recent years, including in 281-282.
IoVT systems. However, edge devices often lack [13] Z. Hu, S. Gnatyuk, T. Okhrimenko, S. Tynymbayev, and M. Iavich,
computational capabilities (processing power, memory) to “"High-Speed and Secure PRNG for Cryptographic Applications," Int.
guarantee sufficient performance (e.g., accuracy, delay, J. Comput. Netw. Inf. Secur, vol. 3, pp. 1–10, 2020
throughput) for complex real-time video analytics tasks. To [14] T. Adame, A. Bel and B. Bellalta, "Increasing LPWAN Scalability by
address this critical issue, the prevailing trend is to offload Means of Concurrent Multiband IoT Technologies: An Industry 4.0
Use Case," in IEEE Access, vol. 7, pp. 46990-47010, 2019.
some video analytics tasks from edge devices to the cloud.
However, existing offloading approaches do not take into [15] F. Meneghello, et al., "IoT: Internet of Threats? A Survey of Practical
Security Vulnerabilities in Real IoT Devices," in IEEE Internet of
account the dynamic nature of video analytics tasks (e.g., Things Journal, vol. 6, no. 5, pp. 8182-8201, Oct. 2019.
different task priorities) and are unable to adapt to system [16] A. Poniszewska-Maranda, R. Matusiak, N. Kryvinska, Ansar-Ul-
dynamics (e.g., different workloads between edge and cloud). Haque Yasar, “A Real-time Service System in the Cloud”, Springer,
To overcome limitations of existing approaches, we develop Journal of Ambient Intelligence and Humanized Computing, 11, pp.
an SDN-based IoVT platform and algorithm for the dynamic 961–977 (2020).
selection of Edge/Cloud computing. The algorithm considers [17] N. Kryvinska, Do van Thanh, "Enabling Voice Services in
such parameters as the priority of computing tasks, the number Hybrid/Cross Network Framework", Proceedings of the London
Communications Symposium 2006, London, UK, 2006, pp. 57-60.
of video streams, and image quality. It can be adapted to a
[18] N. Kryvinska, P. Zinterhof, Do van Thanh, "An Analytical Approach
certain application by software customization of the IoVT to the Efficient Real-time Events/Services Handling in Converged
platform. Significant benchmarking was conducted, which Network Environment", the First International Conference on
will serve as the basis for our future experiments on real-time Network-Based Information Systems, September 3, 2007, pp. 308-316.
video analytics for artificial intelligence applications. The [19] M. Beshley, M. Klymash, M. Hamal, Y. Shkoropad and A. Branytskyy,
experimental results show that our algorithm selects the "Method for Estimating Service Delay in Edge and Cloud Computing
optimal Edge/Cloud computing resources to support real-time Architecture," 2020 IEEE 15th International Conference on Advanced
Trends in Radioelectronics, Telecommunications and Computer
video analytics. In addition, the algorithm proved to be stable Engineering (TCSET), Lviv-Slavske, Ukraine, 2020, pp. 915-919.
under different system loads.
[20] M. Beshley, M. Seliuchenko, O. Panchenko, O. Zyuzko and I. Kahalo,
"Experimental performance analysis of software-defined network
REFERENCES switch and controller," 2018 14th International Conference on
[1] A. Poniszewska-Maranda, D. Kaczmarek, N. Kryvinska, F. Xhafa, Advanced Trends in Radioelecrtronics, Telecommunications and
“Studying usability of AI in the IoT systems/paradigm through Computer Engineering (TCSET), 2018, pp. 282-286.
embedding NN techniques into mobile smart service system”, Springer [21] S. Siddiqui et al., "Toward Software-Defined Networking-Based IoT
Journal, Computing, November 2019, vol. 101, no. 11, pp. 1661–1685. Frameworks: A Systematic Literature Review, Taxonomy, Open
[2] C. W. Chen, "Internet of Video Things: Next-Generation IoT with Challenges and Prospects," IEEE Access, vol. 10, pp. 70850-70901,
Visual Sensors," in IEEE Internet of Things Journal, vol. 7, no. 8, pp. 2022.
6676-6685, Aug. 2020. [22] O. Panchenko, A. Polishuk, M. Seliuchenko and M. Beshley, "Method
[3] M. Seliuchenko, M. Beshley, M. Kyryk and M. Zhovtonoh, for adaptive client oriented management of quality of service in
"Automated Recovery of Server Applications for SDN-Based Internet integrated SDN/CLOUD networks," 2017 4th International Scientific-
of Things," 2019 3rd International Conference on Advanced Practical Conference Problems of Infocommunications. Science and
Information and Communications Technologies (AICT), Lviv, Technology (PIC S&T), Kharkov, Ukraine, 2017, pp. 452-455.
Ukraine, 2019, pp. 149-152. [23] R. Muñoz et al., "Integration of IoT, Transport SDN, and Edge/Cloud
[4] A. Mohan, K. Gauen, Y. -H. Lu, W. W. Li and X. Chen, "Internet of Computing for Dynamic Distribution of IoT Analytics and Efficient
video things in 2030: A world with many cameras," 2017 IEEE Use of Network Resources," in Journal of Lightwave Technology, vol.
International Symposium on Circuits and Systems (ISCAS), Baltimore, 36, no. 7, pp. 1420-1428, 1 April1, 2018.
MD, USA, 2017, pp. 1-4. [24] M. Grega, A. Matiolański, P. Guzik, and M. Leszczuk, “Automated
[5] T.-H. Kim, C. Ramos, and S. Mohammed, “Smart city and IoT,” Future detection of firearms and knives in a CCTV image,” Sensors (Basel),
Gener. Comput. Syst., vol. 76, pp. 159–162, 2017. vol. 16, no. 1, p. 47, 2016.
[6] J. Barthélemy, N. Verstaevel, H. Forehead, and P. Perez, “Edge- [25] M. Beshley, N. Kryvinska, H. Beshley, O. Kochan, and L. Barolli,
computing video analytics for real-time traffic monitoring in a smart "Measuring End-to-End Delay in Low Energy SDN IoT Platform,"
city,” Sensors (Basel), vol. 19, no. 9, p. 2048, 2019. Computers, Materials & Continua, vol. 70, no. 1, pp. 19–41, 2021.
[7] Z. Hu, Y. V. Bodyanskiy, N. Ye. Kulishova, and O. K. Tyshchenko, [26] J. Zhang, X. Zhou, T. Ge, X. Wang and T. Hwang, "Joint Task
“A Multidimensional Extended Neo-Fuzzy Neuron for Facial Scheduling and Containerizing for Efficient Edge Computing,"
Expression Recognition,” International Journal of Intelligent Systems in IEEE Transactions on Parallel and Distributed Systems, vol. 32, no.
and Applications, vol. 9, no. 9, pp. 29–36, Sep. 2017. 8, pp. 2086-2100, 1 Aug. 2021.
[8] M. Fang, et al. “Examination of Abnormal Behavior Detection Based [27] W. Chang, Y. Xiao, W. Lou and G. Shou, "Offloading Decision in Edge
on Improved YOLOv3,” Electronics, vol. 10, no. 2, p. 197, Jan. 2021. Computing for Continuous Applications Under Uncertainty," in IEEE
[9] J. Redmon and A. Farhadi, “YOLOv3: An Incremental Improvement,” Trans. Wirel. Commun., vol. 19, no. 9, pp. 6196-6209, 2020.
arXiv [cs.CV], 2018. [28] N. Kryvinska. Intelligent Network Analysis by Closed Queuing
[10] V. Romanchuk, M. Beshley, A. Polishuk and M. Seliuchenko, "Method Models. Telecommunication Systems 27, 85–98 (2004).
for processing multiservice traffic in network node based on adaptive [29] Z. Hu, et al. “Video Shots Matching via Various Length of
management of buffer resource," 2018 14th International Conference Multidimensional Time Sequences,” International Journal of
on Advanced Trends in Radioelecrtronics, Telecommunications and Intelligent Systems and Applications, vol. 9, no. 11, pp. 10–16, 2017.

Authorized licensed use limited to: Universidad Peruana de Ciencias Aplicadas (UPC). Downloaded on April 15,2023 at 19:05:31 UTC from IEEE Xplore. Restrictions apply.