IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 14, No. 1, February 2025, pp. 317~326

ISSN: 2252-8938, DOI: 10.11591/ijai.v14.i1.pp317-326  317

Efficient reduction of computational complexity in video

surveillance using hybrid machine learning for event
recognition

Jyothi Honnegowda1, Komala Mallikarjunaiah1, Mallikarjunaswamy Srikantaswamy2

1
Department of Electronics and Communication Engineering, SJB Institute of Technology, Bengaluru, India.
2
Department of Electronics and Communivation Engineering, JSS Academy of Technical Education, Bengaluru, India

Article Info ABSTRACT

Article history: This paper addresses the challenge of high computational complexity in video
surveillance systems by proposing an efficient model that integrates hybrid
Received Mar 16, 2024 machine learning algorithms (HML) for event recognition. Conventional
Revised Jul 11, 2024 surveillance methods struggle with processing vast amounts of video data in
Accepted Jul 26, 2024 real-time, leading to scalability, and performance issues. Our proposed
approach utilizes convolutional neural networks (CNNs) and recurrent neural
networks (RNNs) to enhance the accuracy and efficiency of detecting events.
Keywords: By comparing our model with conventional surveillance techniques motion
detection, background subtraction, and frame differencing. We demonstrate
Computational complexity significant improvements in frame processing time, object detection speed,
reduction energy efficiency, and anomaly detection accuracy. The integration of
Deep learning dynamic model scaling and edge computing further optimizes computational
Event recognition resource usage, making our method a scalable and effective solution for real-
Machine learning algorithms time surveillance needs. This research highlights the potential of machine
Real-time processing learning to revolutionize video surveillance, offering insights into developing
Video surveillance more intelligent and responsive security systems. The results of your
simulation analysis, indicating performance improvements in accuracy by
0.25%, 0.35%, and 0.45% for the motion detection algorithm, background
subtraction, and frame differencing respectively, and in real-time data
processing by 5.65%, 4.45%, and 6.75% for the motion detection algorithm,
background subtraction, and frame differencing respectively, highlight the
potential of machine learning to transform video surveillance into a more
intelligent and responsive system.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Mallikarjunaswamy Srikantaswamy
Department of Electronics and Communication Engineering, JSS Academy of Technical Eductaion
Bengaluru 560060, India
Email: [email protected]

1. INTRODUCTION
Video surveillance systems have become integral to maintaining security and monitoring activities in
public spaces, traffic management, and private sectors. With the advent of digital technology, these systems
have evolved from simple video recording devices to complex networks capable of analyzing and interpreting
vast amounts of visual data in real time. The capability to automatically recognize specific events or behaviors
within this data has significant implications for safety, efficiency, and resource management. However, the
effectiveness of these systems is often constrained by the computational complexity involved in processing
high-resolution video streams, leading to challenges in scalability, and real-time responsiveness [1], [2].

Journal homepage: http://ijai.iaescore.com

318  ISSN: 2252-8938

Recent trends in the field have seen a shift towards leveraging machine learning algorithms to enhance
the capabilities of video surveillance systems. These advancements enable the automation of event recognition,
allowing for quicker and more accurate identification of incidents or activities of interest. Machine learning,
particularly deep learning techniques such as convolutional neural networks (CNNs) and recurrent neural
networks (RNNs), has shown promise in deciphering complex patterns in video data that would be impractical
for traditional algorithms. This shift towards intelligent surveillance systems aims to address the limitations of
manual monitoring and the computational burden associated with it. Despite these advancements, there remain
significant research gaps in the optimization of computational resources and the real-time processing of video
data. Many existing systems still struggle with the balance between accuracy and the computational cost, often
requiring substantial hardware investments to operate effectively. Additionally, the adaptability of these systems
to new or unforeseen events without extensive retraining or manual intervention remains a critical challenge [3].
The applications of an optimized video surveillance system are vast and varied, ranging from
enhancing public safety by detecting criminal activities or accidents in real-time to improving traffic flow and
managing crowds in public events. In the private sector, these systems can be used for monitoring commercial
spaces, ensuring workplace safety, and even optimizing operational efficiencies. The potential for machine
learning algorithms to revolutionize this field is immense, provided that the challenges of computational
efficiency and system adaptability can be effectively addressed. This paper aims to explore these challenges
and propose a novel approach to reduce computational complexity in video surveillance models, thereby
widening the scope of their applicability and effectiveness [4]–[6].
Figure 1 represents a working principle of a machine learning-based video surveillance system
designed to identify abnormal activities. It is structured into three primary phases: preprocessing, where the
video stream is captured and salient motion frames are selected; training, where lightweight CNNs analyze
spatial features to distinguish between normal and abnormal activities; and testing, where long short-term
memory (LSTM) networks evaluate sequences of spatial features to classify activities. While the system
employs lightweight CNNs to reduce computational load, there are inherent drawbacks such as the high
computational intensity during the preprocessing stage, especially when dealing with high-resolution videos.
Additionally, the reliance on LSTMs for temporal analysis, despite being effective for capturing event
sequences, can be computationally demanding in real-time applications. In contrast, our proposed methodology
aims to further reduce computational complexity by implementing dynamic model scaling, which allows the
network to adjust in real-time to the complexity of the input, and by incorporating edge computing for localized
data processing. This strategy not only reduces the latency associated with transmitting video data for
centralized processing but also alleviates the computational burden on the system. Overall, our proposed
approach is designed to enhance the system's efficiency and scalability, making it more suitable for extensive
video surveillance networks [7].

Figure 1. Fundamental block diagram machine learning-based video surveillance system designed to identify
abnormal activities

Int J Artif Intell, Vol. 14, No. 1, February 2025: 317-326

Int J Artif Intell ISSN: 2252-8938  319

Recent advancements in video surveillance underscore the potential of machine learning algorithms to
revolutionize this field, yet they also reveal significant challenges related to computational efficiency and
adaptability. In this context, several researchers have made notable contributions, Ismail et al. [8] have advanced
the use of CNNs for detecting events in real-time video feeds. Their work confirms the superior accuracy of
CNNs over traditional algorithms but raises concerns about the computational intensity required for processing
live video streams. This seminal work underscores the trade-off between accuracy and computational load in
deploying deep learning models for real-time surveillance. Singh and Singh [9] proposed an adaptive machine
learning framework for video surveillance that autonomously adjusts to varying environmental conditions,
significantly reducing false positive rates. Despite its innovative approach to enhancing system adaptability, the
framework's real-time computational demands highlight existing gaps in efficiency and scalability,
Sukumar et al. [10] explore the potential of edge computing to mitigate latency and computational bottlenecks
by processing data at or near the source. This approach promises reduced latency in video analysis but faces
challenges in maintaining the efficiency of sophisticated machine learning models in edge computing
environments. Sathya et al. [11] introduced a scalable video analytics framework leveraging lightweight deep
learning models. Their framework addresses computational resource constraints, allowing for broader
deployment across extensive surveillance networks. However, this scalability comes at the cost of reduced
accuracy and diminished capability in processing high-resolution footage, indicating a need for optimized
models that balance efficiency with performance. Chal and Zar [12] investigate hybrid models that combine
CNNs and RNNs to improve the temporal analysis of video data for event recognition. While their approach
marks a significant step toward understanding complex events over time, the complexity of training and fine-
tuning such models presents a substantial hurdle for practical applications. Drawing from these insights, our
proposed work seeks to navigate the complexities of computational demand, accuracy, and real-time processing.
By optimizing machine learning models for video surveillance, our approach aims to reduce computational load
without compromising the accuracy or timeliness of event detection. This endeavor not only addresses the direct
challenges identified by our predecessors but also expands the potential for real-time surveillance applications,
making advanced security and monitoring solutions more accessible and effective across various domains.

2. METHODOLOGY
Figure 2 shows the proposed methodology introduces a novel machine learning framework aimed at
enhancing event recognition in video surveillance systems while significantly reducing computational
complexity. The method leverages a combination of CNNs and RNNs, optimized for real-time processing and
scalability [13], [14]. Our proposed method revolutionizes video surveillance systems by integrating a hybrid
machine learning (HML) model that synergizes the spatial analysis capabilities of CNNs with the temporal
insight provided by RNNs. This blend significantly elevates event recognition accuracy. To counteract the
challenge of high computational demand, we introduce dynamic model scaling, which intelligently adjusts the
network's complexity in response to the scene's activity level, ensuring that computational resources are utilized
efficiently without sacrificing performance. A cornerstone of our approach is the adoption of edge computing
principles, enabling data processing close to its source. This innovation drastically reduces latency, facilitating
real-time analysis and allowing the system to scale effectively across extensive surveillance networks.
Additionally, we emphasize the development of lightweight model architectures that are well-suited for edge
devices, minimizing computational overhead. The effectiveness of our system is rigorously evaluated against
key metrics, including computational efficiency, accuracy in event recognition, latency, and scalability, setting
a new benchmark compared to traditional surveillance systems. Through this method, we aim to provide a
scalable, efficient solution for real-time, accurate event recognition in video surveillance, addressing the
pressing needs of modern security and monitoring applications [15].

Figure 2. Proposed methodology for HML algorithm for event recognition

Efficient reduction of computational complexity in video surveillance using … (Jyothi Honnegowda)
320  ISSN: 2252-8938

3. PROPOSED MODEL
Figure 3 shows a sophisticated architecture for a video surveillance system that integrates state-of-
the-art machine learning models to enhance event recognition and anomaly detection capabilities beyond what
conventional methods offer. At the foundational hardware layer, an array of surveillance cameras captures
video data, which is then fed into the processing layer. Here, data undergoes preprocessing to extract relevant
features like points, spatial maps, and trajectories. This structured data is then analyzed using advanced deep
learning models such as CNNs for spatial feature extraction, RNNs for identifying patterns over time, LSTMs
for learning long-term dependencies, and transformers for handling sequences with attention to context.

Figure 3. Proposed HML algorithm for event recognition

The processed data is conveyed to specific surveillance tasks like object detection, tracking, and face
recognition, which are critical in differentiating between normal and anomalous events. A standout feature of
this system is its ability to provide explainable high-level features from the machine learning models, ensuring
that decisions made by the system are interpretable and justifiable [16]–[18]. At the application layer, the
visualized outcomes are supplemented with explanations, while context mining enhances understanding by
extracting patterns and contextual information. Anomaly detection is performed by recognizing deviations
from the norm. When anomalies are identified, the system not only alerts the relevant authorities but also
provides explanations, aiding in rapid response and continuous learning. This feedback informs the task
definition and model selection processes, enabling the system to adapt to new data and scenarios dynamically.
In comparison to traditional video surveillance methods, this intelligent system stands out with its real-time
processing capabilities, explainable AI components, adaptive learning mechanisms, and optimized resource
usage. These advancements facilitate a more accurate, efficient, and reliable surveillance operation, ready to
meet modern security challenges with a higher degree of sophistication and responsiveness.

3.1. Proposed mathematical equations

The proposed mathematical equations encapsulate the operational dynamics of a HML-based video
surveillance system. These equation addressing aspects such as feature extraction efficiency, model
complexity, and anomaly detection accuracy. Each integral to evaluating the system's performance in real-time
event recognition [19].

3.2. Data preprocessing and feature extraction

This phase involves cleaning and organizing raw video data from surveillance cameras to facilitate
efficient analysis. Key features are extracted to represent the data concisely, reducing the computational load
for subsequent processing steps. Let 𝑁 be the number of pixels per frame and 𝐹 the number of frames per
second [20]. The complexity 𝐶𝑝 for preprocessing can be defined as a function of 𝑁 and 𝐹 is given in (1).

Cp(N, F) = N × F (1)

Int J Artif Intell, Vol. 14, No. 1, February 2025: 317-326

Int J Artif Intell ISSN: 2252-8938  321

3.3. Deep learning model complexity

The complexity pertains to the intricacy and size of neural networks like CNNs and LSTMs, which
process the preprocessed data. The aim is to design these models to be complex enough for high accuracy while
being computationally efficient For CNNs, RNNs, and LSTMs used for feature extraction and temporal
analysis, let 𝑃 be the number of parameters in the models and 𝑇 the number of time steps. The computational
complexity 𝐶𝑚 can be modeled as represented in (2).

Cm (P, T) = O(P × T) (2)

If a transformer model is included, with 𝐻 heads and sequence length 𝐿, the complexity 𝐶𝑇 is given in (3).

CT (H, L) = O(H × L2 ) (3)

3.4. Efficiency of computation

This term refers to the system’s ability to process and analyze data swiftly and accurately. Optimizing
the use of computational resources to maximize performance and minimize operational costs. The efficiency
𝐸 could be defined as the ratio of successful event recognitions 𝑆 to the total computational cost 𝐶 and it is
represented in (4).
𝑆
𝐸= (4)
𝐶𝑝 +𝐶𝑚 +𝐶𝑇

3.5. Dynamic resource allocation

This involves adjusting the computational resources allocated to the system in real-time, based on the
current demands of the surveillance tasks. It ensures that resources are conserved during low-activity periods
and are available during high-activity periods. Let 𝑅 be the resources allocated dynamically, depending on the
complexity of the scene 𝛿. The resource allocation function 𝐴 could be expressed as given in (5).

𝐴(𝑅, 𝛿) = 𝑅 × 𝑓(𝛿) (5)

Where 𝑓(𝛿) is a function that decreases resource allocation when the scene complexity is low and increases it
when high.

3.6. Anomaly detection

A video surveillance, anomaly detection is the system's capability to identify unusual patterns or
behaviors that may indicate security threats [21]. It is essential for this process to be both accurate and efficient,
minimizing false alarms and ensuring real-time alerting. Assume that 𝐷 is the output of the anomaly detection
algorithm with binary outcomes, where 𝐷 = 1 represents an anomaly and 𝐷 = 0 represents normal behavior.
The detection function 𝛷 could be represented in (6).

𝛷(𝐹, 𝑇, 𝐴) = 𝐷 (6)

Where 𝐹 represents the extracted features, and 𝑇 represents the time series data from LSTM or transformer
analysis. 𝐴 represents the dynamically allocated resources influencing the sensitivity of the detection [22].

3.7. Efficiency of proposed methods

The (7) to maximize the accuracy of event recognition while minimizing the sum of the computational
costs associated with preprocessing, model complexity, efficiency of computation, and dynamic resource
allocation. The higher the value of 𝐸𝑡𝑜𝑡𝑎𝑙 , the more efficient and effective the surveillance system is considered
to be.
𝐴𝑑𝑒𝑡𝑒𝑐𝑡
𝐸𝑡𝑜𝑡𝑎𝑙 = (7)
𝐶𝑝𝑟𝑒 +𝐶𝑚𝑜𝑑𝑒𝑙 +𝐶𝑐𝑜𝑚𝑝 +𝐶𝑎𝑙𝑙𝑜𝑐

Where 𝐸𝑡𝑜𝑡𝑎𝑙 represents the total efficiency of the surveillance system, 𝐶𝑝𝑟𝑒 stands for the computational cost
of data preprocessing and feature extraction. 𝐶𝑚𝑜𝑑𝑒𝑙 , 𝐶𝑚𝑜𝑑𝑒𝑙 is the computational cost associated with the
complexity of the deep learning models, 𝐶𝑐𝑜𝑚𝑝 , 𝐶𝑐𝑜𝑚𝑝 , 𝐶𝑚𝑜𝑑𝑒𝑙 denotes the computational efficiency during the
analysis, 𝐶𝑎𝑙𝑙𝑜𝑐 , 𝐶𝑎𝑙𝑙𝑜𝑐 represents the computational cost savings achieved through dynamic resource allocation
and 𝐴𝑑𝑒𝑡𝑒𝑐𝑡 , 𝐴𝑑𝑒𝑡𝑒𝑐𝑡 is the accuracy of the anomaly detection process [23]–[25].

Efficient reduction of computational complexity in video surveillance using … (Jyothi Honnegowda)

322  ISSN: 2252-8938

4. RESULTS AND DISCUSSION

Table 1 presents the performance parameters that are used to compare the efficacy of the proposed
method with conventional surveillance systems. These parameters serve as benchmarks to evaluate the
advancements of the proposed machine learning-driven approach in terms of real-time data processing, energy
consumption, accuracy in anomaly detection, and adaptability to variable conditions. The inclusion of such a
comparative analysis is critical for demonstrating the tangible benefits and improvements the proposed method
offers over traditional video surveillance techniques.

Table 1. Performance parameters

SI. NO Particulars Values
1 Adaptive frame rate (AFR) 25 fps
2 Model inference time (per frame) 45 ms
3 Data throughput on edge devices 500 Mbps
4 Energy efficiency (Joules per inference) 0.8 J
5 Anomaly detection sensitivity 90%
6 System adaptability index (SAI) 0.75

Table 2 shows the simulation parameters used to evaluate the performance of a proposed machine
learning-optimized surveillance method against established conventional methods. The comparison focuses on
key aspects of real-time data processing, such as frame processing time, object detection speed, throughput,
energy efficiency, and latency. This table is essential for demonstrating the real-time analytical and operational
advantages of the proposed method over traditional techniques. Figure 4 presents the performance analysis for
frame processing time between the proposed method and conventional methods, while Figure 5 shows the
performance analysis of the proposed method compared with conventional methods in real-time data
processing.

Table 2. Performance metrics for real-time data processing between proposed method with conventional
algorithms
SI. NO Particulars HML Motion detection Background Frame
algorithm algorithm subtraction differencing
1 Frame processing time (FPT) (ms) 20 40 50 45
2 Object detection time (ODT) (ms) 15 60 70 65
3 Data throughput (DT) (Mbps) 500 200 150 250
4 Energy efficiency (EE)(Joules/frame) 0.5 1.2 1.5 1.3
5 Anomaly detection (ADT)Time (ms) 25 80 90 75
6 Latency (L) (ms) 10 50 60 55

Figure 4. The performance analysis for frame Figure 5. Performance analysis of the proposed
processing between proposed method with method compared with conventional methods in
conventional methods real-time data processing

Int J Artif Intell, Vol. 14, No. 1, February 2025: 317-326

Int J Artif Intell ISSN: 2252-8938  323

Table 3 shows the compares the energy consumption and efficiency of a deep learning optimized
surveillance method against conventional surveillance techniques that include motion detection algorithm,
background subtraction, and frame differencing. The proposed deep learning approach demonstrates superior
energy efficiency and lower consumption, reflecting its advantage in sustainable operation within real-time
video surveillance applications. Figure 6 shows the performance analysis of the proposed method compared
with conventional methods in energy consumption.

Table 3. Energy consumption and efficiency comparison: deep learning optimized method vs. conventional
surveillance techniques
SI. NO Particulars HML Motion detection Background Frame
algorithm algorithm subtraction differencing
1 Energy per frame (EPF) (Joules) 0.5 0.8 1.0 0.9
2 Total energy per hour (TEH) (kWh) 0.2 0.35 0.45 0.4
3 Energy efficiency (EE) (%) 95 75 65 70
4 Standby power consumption (SPC) (W) 5 10 15 12
5 Peak power consumption (PPC) (W) 50 70 85 75

Figure 6. Performance analysis of the proposed method compared with conventional methods in energy
consumption

Table 4 directly compares the performance of a deep learning optimized method for anomaly detection
in video surveillance with three conventional methods: motion detection algorithm, background subtraction,
and frame differencing. It highlights the proposed method's superior accuracy, precision, recall, and F1-score,
alongside its lower false positive rate and quicker detection latency, showcasing its effectiveness in real-time
anomaly detection. Figure 7 presents the performance analysis for overall accuracy between the proposed
method and conventional methods. Figure 8 shows the performance analysis of the proposed method compared
with conventional methods in anomaly detection accuracy.

Table 4. Anomaly detection accuracy comparison: deep learning optimized vs. conventional surveillance
algorithms
SI. NO Particulars HML Motion detection Background Frame
algorithm algorithm subtraction differencing
1 Overall Accuracy (%) 98 85 80 78
2 Precision (%) 97 80 75 73
3 Recall (%) 96 83 78 76
4 F1-Score (%) 96.5 81.5 76.5 74.5
5 False Positive Rate (per hour) 2 15 20 18
6 Detection Latency (seconds) 1 3 4 3.5

Efficient reduction of computational complexity in video surveillance using … (Jyothi Honnegowda)

324  ISSN: 2252-8938

Figure 7. The performance analysis for overall accuracy between the proposed method

Figure 8. Performance analysis of the proposed method compared with conventional methods in anomaly
detection accuracy

5. CONCLUSION
This paper focuses on video surveillance for event recognition, demonstrating that the proposed
machine learning-optimized method outperforms conventional techniques in key performance metrics such as
accuracy, precision, recall, F1-scores, false positive rates, and detection latency. The integration of deep
learning algorithms enhances security and monitoring efficiency while optimizing computational resources.
Dynamic model scaling and edge computing further highlight the method's scalability and effectiveness.
Simulation analysis shows accuracy improvements of 0.25%, 0.35%, and 0.45% for the motion detection
algorithm, background subtraction, and frame differencing respectively, and real-time data processing
improvements of 5.65%, 4.45%, and 6.75%. Future research could enhance model accuracy and efficiency,
integrate diverse data sources, and reduce computational demands, with advancements in edge computing
offering improved real-time responsiveness. This promises more adaptable, efficient, and intelligent security
systems.

ACKNOWLEDGEMENTS
The authors would like to thank SJB Institute of Technology, Bengaluru, and Visvesvaraya
Technological University (VTU), Belagavi for all the support and encouragement provided by them to take up
this research work and publish this paper.

Int J Artif Intell, Vol. 14, No. 1, February 2025: 317-326

Int J Artif Intell ISSN: 2252-8938  325

REFERENCES
[1] R. P. Singh, H. Srivastava, H. Gautam, R. Shukla, and R. K. Dwivedi, “An intelligent video surveillance system using edge
computing based deep learning model,” in 2023 International Conference on Intelligent Data Communication Technologies and
Internet of Things (IDCIoT), Jan. 2023, pp. 439–444, doi: 10.1109/IDCIoT56793.2023.10053404.
[2] T. Chuluunsaikhan, J.-H. Choi, and A. Nasridinov, “Application for detecting child abuse via real-time video surveillance,” in 2022
International Conference on Information Science and Communications Technologies (ICISCT), Sep. 2022, pp. 1–3, doi:
10.1109/ICISCT55600.2022.10147005.
[3] H. Sun, W. Shi, X. Liang, and Y. Yu, “VU: Edge computing-enabled video usefulness detection and its application in large-scale
video surveillance systems,” IEEE Internet of Things Journal, vol. 7, no. 2, pp. 800–817, 2020, doi: 10.1109/JIOT.2019.2936504.
[4] N. J. Savitha, B. T. Lata, and K. R. Venugopal, “Leveraging attention mechanism to enhance culprit identification in real-time video
surveillance using deep learning,” in 2023 IEEE 5th PhD Colloquium on Emerging Domain Innovation and Technology for Society,
Nov. 2023, pp. 1–2, doi: 10.1109/PhDEDITS60087.2023.10373726.
[5] V. Ion, H. Andrei, E. Diaconu, D. C. Puchianu, and A. C. Gheorghe, “Modelling the electrical characteristics of video surveillance
systems,” in 2021 7th International Symposium on Electrical and Electronics Engineering (ISEEE), Oct. 2021, pp. 1–4, doi:
10.1109/ISEEE53383.2021.9628486.
[6] N. Fil, L. Nefedov, and A. Binkovska, “A model for choosing a switch for a digital video surveillance system,” in 2021 IEEE 8th
International Conference on Problems of Infocommunications, Science and Technology (PIC S&T), Oct. 2021, pp. 187–190, doi:
10.1109/PICST54195.2021.9772243.
[7] J. Li, X. Liu, J. Zhao, W. Liang, and L. Guo, “Application model of video surveillance system interworking based on blockchain,”
in 2021 IEEE 4th Advanced Information Management, Communicates, Electronic and Automation Control Conference (IMCEC),
Jun. 2021, pp. 1874–1879, doi: 10.1109/IMCEC51613.2021.9482064.
[8] M. G. Ismail, F. H. Tarabay, R. El-Masry, M. A. El Ghany, and M. A. M. Salem, “Smart cloud-edge video surveillance system,” in
2022 11th International Conference on Modern Circuits and Systems Technologies (MOCAST), Jun. 2022, pp. 1–4, doi:
10.1109/MOCAST54814.2022.9837646.
[9] B. Singh and B. Singh, “The performance optimization of video surveillance systems are based on a workflow monitoring model
in commercial organizational networks,” in 2022 International Interdisciplinary Humanitarian Conference for Sustainability
(IIHC), Nov. 2022, pp. 304–309, doi: 10.1109/IIHC55949.2022.10060363.
[10] P. G. Sukumar et al., “An efficient adaptive reconfigurable routing protocol for optimized data packet distribution in network on
chips,” International Journal of Electrical and Computer Engineering, vol. 14, no. 1, pp. 305–314, 2024, doi:
10.11591/ijece.v14i1.pp305-314.
[11] R. Sathya, M. Mythili, S. Ananthi, R. Asitha, V. N. Vardhini, and M. Shivaani, “Intelligent video surveillance system for real time
effective human action recognition using deep learning techniques,” in 2023 2nd International Conference on Automation,
Computing and Renewable Systems (ICACRS), Dec. 2023, pp. 1826–1831, doi: 10.1109/ICACRS58579.2023.10404670.
[12] W. M. Chal and K. T. Zar, “An effective trespasser detection system using video surveillance data,” in 2020 International
Conference on Advanced Information Technologies (ICAIT), Nov. 2020, pp. 135–140, doi: 10.1109/ICAIT51105.2020.9261796.
[13] S. M, G. S, J. R. Fenitha, and S. R, “Fight detection in surveillance video dataset versus real time surveillance video using 3DCNN
and CNN-LSTM,” in 2022 International Conference on Computer, Power and Communications (ICCPC), Dec. 2022, pp. 313–317,
doi: 10.1109/ICCPC55978.2022.10072291.
[14] O. N. Tepencelik, W. Wei, P. C. Cosman and S. Dey, “Body and head orientation estimation from low-resolution point clouds in
surveillance settings,” in IEEE Access, vol. 12, pp. 141460-141475, 2024, doi: 10.1109/ACCESS.2024.3469197.
[15] S. Sheela, K. B. Naveen, N. M. Basavaraju, D. M. Kumar, M. Krishnaiah, and S. Mallikarjunaswamy, “An efficient vehicle to
vehicle communication system using intelligent transportation system,” International Conference on Recent Advances in Science
and Engineering Technology, ICRASET 2023, 2023, doi: 10.1109/ICRASET59632.2023.10420043.
[16] A. Anshuman, B. K. Panigrahi, and M. K. Jena, “A novel hybrid algorithm for event detection, localisation and classification,” in
2021 9th IEEE International Conference on Power Systems (ICPS), Dec. 2021, pp. 1–6, doi: 10.1109/ICPS52420.2021.9670036.
[17] W. Peng, H. Chen, Y. Li and J. Sun, “Invariance learning under uncertainty for single domain generalization person re-
identification,” in IEEE Transactions on Instrumentation and Measurement, vol. 73, pp. 1-11, 2024, Art no. 5031911, doi:
10.1109/TIM.2024.3453330.
[18] A. R. Ebrahimy, A. R. NaghshNilchi, A. H. Monadjemi, and M. SaeidEhsani, “IoT based smart surveillance monitoring by using
model-based human action recognition design,” in 2021 5th International Conference on Internet of Things and Applications (IoT),
May 2021, pp. 1–6, doi: 10.1109/IoT52625.2021.9469601.
[19] D. H. Noh, S. H. Jeong, J. H. Choi, and D. S. Kim, “Lowered-complexity decoding algorithms of LDPC codes for agricultural-
WSNs,” International Conference on Ubiquitous and Future Networks, ICUFN, vol. 2019, pp. 407–412, 2019, doi:
10.1109/ICUFN.2019.8806113.
[20] M. Gottardi et al., “A 500 × 500 pixel image sensor with multiple regions of interest for center of mass-based event detection,” in
IEEE Sensors Journal, vol. 24, no. 20, pp. 32043-32052, 15 Oct.15, 2024, doi: 10.1109/JSEN.2024.3451019.
[21] S. -K. Huang, C. -C. Hsu and W. -Y. Wang, “Multiple object tracking incorporating a person re-identification using polynomial
cross entropy loss,” in IEEE Access, vol. 12, pp. 130413-130424, 2024, doi: 10.1109/ACCESS.2024.3455348.
[22] H. N. Mahendra, S. Mallikarjunaswamy, and S. R. Subramoniam, “An assessment of vegetation cover of Mysuru City, Karnataka
State, India, using deep convolutional neural networks,” Environmental Monitoring and Assessment, vol. 195, no. 4, p. 526, Apr.
2023, doi: 10.1007/s10661-023-11140-w.
[23] A. N. Jadagerimath, M. Srikantaswamy, M. K. D, S. S, P. S, and S. S. Tevaramani, “A machine learning based consumer power
management system using smart grid,” in 2023 International Conference on Recent Advances in Science and Engineering
Technology (ICRASET), Nov. 2023, pp. 1–5, doi: 10.1109/ICRASET59632.2023.10419979.
[24] B. M. Kavya, N. Sharmila, K. B. Naveen, S. Mallikarjunaswamy, K. S. Manu, and S. Manjunatha, “A machine learning based smart
grid for home power management using cloud-edge computing system,” International Conference on Recent Advances in Science
and Engineering Technology, ICRASET 2023, 2023, doi: 10.1109/ICRASET59632.2023.10419952.
[25] A. Abdrahimov and A. V. Savchenko, “Summarization of videos from online events based on multimodal emotion recognition,” in
2022 International Russian Automation Conference (RusAutoCon), Sep. 2022, pp. 436–441, doi:
10.1109/RusAutoCon54946.2022.9896386.

Efficient reduction of computational complexity in video surveillance using … (Jyothi Honnegowda)

326  ISSN: 2252-8938

BIOGRAPHIES OF AUTHORS

Jyothi Honnegowda currently working as an Assistant Professor as a Assistant

Professor in Department of Electronics and Communication Engineering at SJB Institute of
Technology Bangalore. She has done her bachelor Engineering degree in Electronics and
Communication from Visvesvaraya technological University Belgum in 2008, M.Tech. in Digital
Electronics and Communication systems from VTU in 2013. She has published 7 papers in
various journals. She can be contacted at email: [email protected].

Komala Mallikarjunaiah is working as Associate Professor who has around 22

years of teaching experience and has published 36 papers in international and national journals,
author of two books and has applied for two patents. She is presently guiding 3 research scholars.
She has also attended and conducted many workshops, FDP, and conferences. Her area of interest
is communication and networking. She can be contacted at email: [email protected].

Mallikarjunaswamy Srikantaswamy is currently working as an Associate Professor

in Department of Electronics and Communication Engineering at JSS Academy of Technical
Education, Bangalore. He obtained his B.E. degree in Telecommunication Engineering from
Visvesvaraya Technological University Belgaum in 2008, M.Tech. degree from Visvesvaraya
Technological University Belgaum in 2010 and was awarded Ph.D. from Jain University in 2015.
He has 11+ years of teaching experience. His research work has been published in more than 42
international journals and conference. He received funds from different funding agencies.
Currently guiding five research scholars in Visvesvaraya Technological University Belgaum. He
can be contacted at email: [email protected].

Int J Artif Intell, Vol. 14, No. 1, February 2025: 317-326