Optimizing Hybrid Long-Term Recurrent Convolutional Network Architecture for Edge-Based Violence Detection in Video Surveillance Systems
https://doi.org/10.37493/2307-910X.2025.3.3
Abstract
Introduction. This study presents a solution for automated detection of violent incidents in video streams using a hybrid architecture that integrates Convolutional Neural Networks (CNNs) for spatial frame analysis and Long Short-Term Memory (LSTM) networks for identifying temporal dependencies. Materials and methods. The system is optimized for deployment on low-power NVIDIA Jetson series devices, enabling on-edge data processing at the capture site. Experiments were conducted on a specialized dataset comprising urban surveillance footage, sports broadcasts, and simulated scenes. Results and discussion. Results confirmed high recognition accuracy, low falsepositive rates, and minimal processing latency, meeting real-time security system requirements. Operational modes include frame-by-frame processing with timestamp annotation and file-based analysis for rapid video assessment. Conclusion. Integration with security infrastructure and content moderation platforms represents a key implementation pathway
About the Authors
V. M. Goryaev
Kalmyk State University named after B. B. Gorodovikov
Россия
Россия
Goryaev Vladimir Mikhailovich, Candidate of Pedagogical Sciences, Associate Professor
11, Pushkin St., Elista
S. A. Mankaeva
Kalmyk State University named after B. B. Gorodovikov
Россия
Россия
Mankaeva Saglar Alekseevna, 2nd year student
phone number: +79961102363
11, Pushkin St., Elista
E. V. Sumyanova
Kalmyk State University named after B. B. Gorodovikov
Россия
Россия
Sumyanova Elena Vladimirovna, Associate Professor of the Department of Experimental and General Physics
+79615486561
11, Pushkin St., Elista
J. B. Bembitov
Kalmyk State University named after B. B. Gorodovikov
Россия
Россия
Bembitov Jirgal Batrovich, Associate Professor of the Department of Theoretical Physics
+79615400560
11, Pushkin St., Elista
G. A. Mankaeva
Kalmyk State University named after B. B. Gorodovikov
Россия
Россия
Mankaeva Galina Alekseevna, Senior Lecturer at the Department of Theoretical Physics
phone number: +79061764200
11, Pushkin St., Elista
References
1. Glossarii terminov po informatike, vychislitel'noi tekhnike i komp'yuternym setyam / Pod red. I.A. Sokolova. M.: IKSI RAN. 2023. 214 s.
2. Donahue J., Hendricks L. et al. LRCN for Visual Recognition and Description // IEEE Transactions on Pattern Analysis and Machine Intelligence. 2016. Vol. 39. Is. 4. P. 677–691.
3. Goryaev V.M., Matsakov B.V. Issledovanie ehffektivnosti modelei neiroseti LRCN v zadachakh raspoznavaniya nasiliya na video//Sovremennye naukoemkie tekhnologii. 2024. № 12. S. 17-24 DOI: https://doi.org/10.17513/snt.40239
4. Sudhakaran S., Lanz O. Learning to Detect Violent Videos using Convolutional Long ShortTerm Memory // IEEE International Conference on Computer Vision Workshops (ICCVW). 2019. P. 1– 9. DOI: 10.1109/ICCVW.2019.00010
5. Bilinski P., Bremond F. Human Violence Recognition and Detection in Surveillance Videos // IEEE International Conference on Advanced Video and Signal-Based Surveillance (AVSS). 2020. P. 1–8. DOI: 10.1109/AVSS.2020.9339832
6. Matsakov B.V., Goryaev V.M. Raspoznavanie priznakov nasiliya s pomoshch'yu setei dolgoi kratkosrochnoi pamyati LSTM: Svidetel'stvo o gosudarstvennoi registratsii programmy dlya EHVM № 2024690113. Zayavka № 2024688048 ot 15.11.2024. Zaregistrirovano 12.12.2024.
7. Hochreiter S., Schmidhuber J. Long Short-Term Memory // Neural Computation. 1997. Vol. 9. Is. 8. P. 1735–1780. DOI: 10.1162/neco.1997.9.8.1735
8. Pustynnyi YA.N. Reshenie problemy ischezayushchego gradienta s pomoshch'yu neironnykh setei dolgoi kratkosrochnoi pamyati. Innovatsii i investitsii. 2020. №2. C.130-132.
9. Wang L.et al. Temporal segment networks: Towards good practices for deep action recognition // Proceedings of the European Conference on Computer Vision (ECCV 2016). Amsterdam. Oct. 8–16, 2016. Cham: Springer. 2016. P. 20–36. DOI: 10.1007/978-3-319-46484-8_29.
10. NVIDIA Jetson Xavier NX Developer Guide. 2023. URL: https://developer.nvidia.com/embedded/jetson-xavier-nx (data obrashcheniya: 20.06.2025).
11. Hanson A., Pnvr K., Krishnagopal S., Davis L. Bidirectional Convolutional LSTM for the Detection of Violence in Videos // IEEE Transactions on Cognitive and Developmental Systems. 2021. Vol. 13. Is. 4. P. 992–1001. DOI: 10.1109/TCDS.2020.2996775
12. Zhang H. et al. Edge Computing for Real-Time Video Analysis // IEEE Internet of Things Journal. 2021. Vol. 8. Is. 16. P. 12539–12551. DOI: 10.1109/JIOT.2021.3065956
13. Petrov K.L. Ehvolyutsiya paradigmy periferiinykh vychislenii // Informatsionnye tekhnologii. 2022. T.28. №4. S.12-19.DOI: 10.22213/2410-9304-2022-4-12-19
14. Google Content Safety API. Developer Documentation. [Ehlektronnyi resurs]. 2023. URL: https://cloud.google.com/content-safety/docs (data obrashcheniya: 20.06.2025).
15. ViSenze. Real-Time Violence Detection in Video Streams [Ehlektronnyi resurs]. 2024. URL: https://www.visenze.com/resources (data obrashcheniya: 20.06.2025).
16. Redmon J., Farhadi A. YOLOv3: An Incremental Improvement // IEEE Transactions on Pattern Analysis and Machine Intelligence. 2020. Vol. 42. Is. 2. P. 1–14.
17. GOST R ISO 5725-2002. Tochnost' metodov i rezul'tatov izmerenii. M.: Standartinform, 2002. 31 c. [Текст]
18. Lyubchich V. et al. A distribution-free m-out-of-n bootstrap approach to testing symmetry about an unknown median. Computational Statistics & Data Analysis.2016.Vol.104.P.1-9. DOI: 10.1016/j.csda.2016.05.004
Review
For citations:
Goryaev V.M., Mankaeva S.A., Sumyanova E.V., Bembitov J.B., Mankaeva G.A. Optimizing Hybrid Long-Term Recurrent Convolutional Network Architecture for Edge-Based Violence Detection in Video Surveillance Systems. Modern Science and Innovations. 2025;(3):30-38. (In Russ.) https://doi.org/10.37493/2307-910X.2025.3.3
Views:
15
JATS XML
Email this article 