The article analyzes the architecture of intelligent video surveillance based on an ontology of information quanta. The relevance of the study is determined by tightening requirements for IoT video analytics, which must operate under limited network bandwidth, strict privacy requirements, and tight budgets. The novelty of the approach lies in tokenizing video streams at the edge : low-level visual descriptors are transformed into semantically stable information quanta (IQ), after which their cloud processing is performed by vision-language models. The paper formalizes the principles of a two-tier edge-cloud architecture and analyzes data-centric methods that increase the robustness of models to noisy data. Special attention within the work is paid to the ontology of actions as a connecting link between pose detection and subsequent semantic interpretation. The aim of the study is to demonstrate that the proposed architecture ensures ultra-low latency, compliance with privacy requirements, and high interpretability of decisions. To achieve this aim, methods of systems analysis and comparative performance analysis are employed. In conclusion, it is shown that the architecture enables efficient solving of complex event analysis tasks in real time. The material is addressed to specialists in the fields of computer vision, the Internet of Things, and security systems.

Internet of Things, edge-cloud architecture, data-centric AI, ontology of actions, information quantum, vision-language control, low latency, pose recognition, video surveillance

Silva, J., Marques, E. R., Lopes, L. M., & Silva, F. (2021). Energy-aware adaptive offloading of soft real-time jobs in mobile edge clouds. Journal of Cloud Computing, 10(1), 38.

Gan, Z., Li, L., Li, C., Wang, L., Liu, Z., & Gao, J. (2022). Vision-language pre-training: Basics, recent advances, and future trends. Foundations and Trends® in Computer Graphics and Vision, 14(3–4), 163-352. http://dx.doi.org/10.1561/0600000105

Hamid, O. H. (2023). Data-centric and model-centric AI: Twin drivers of compact and robust industry 4.0 solutions. Applied Sciences, 13(5), 2753.

Abu Tami, M., Ashqar, H. I., Elhenawy, M., Glaser, S., & Rakotonirainy, A. (2024). Using Multimodal Large Language Models (MLLMs) for Automated Detection of Traffic Safety-Critical Events. Vehicles, 6(3), 1571-1590. https://doi.org/10.3390/vehicles6030074

Huang, A. Y., Chen, Y., Huang, D., & Zhao, M. (2023, December). Semantic Privacy-Preserving for Video Surveillance Services on the Edge. In Proceedings of the Eighth ACM/IEEE Symposium on Edge Computing (pp. 300-305). https://doi.org/10.1145/3583740.3626820

Diraco, G., Rescio, G., Caroppo, A., Manni, A., & Leone, A. (2023). Human Action Recognition in Smart Living Services and Applications: Context Awareness, Data Availability, Personalization, and Privacy. Sensors, 23(13), 6040. https://doi.org/10.3390/s23136040

Md Nur Hasan Mamun (2024). INTEGRATION OF ARTIFICIAL INTELLIGENCE AND DEVOPS IN SCALABLE AND AGILE PRODUCT DEVELOPMENT: A SYSTEMATIC LITERATURE REVIEW ON FRAMEWORKS. ASRC Procedia: Global Perspectives in Science and Scholarship, 4(1), 01–32. https://doi.org/10.63125/exyqj773

Zha, D., Bhat, Z. P., Lai, K. H., Yang, F., Jiang, Z., Zhong, S., & Hu, X. (2025). Data-centric artificial intelligence: A survey. ACM Computing Surveys, 57(5), 1-42. https://doi.org/10.1145/3711118

Cho, W., Kim, D., Lim, B., & Gu, J. (2025). PreEdgeDB: A Lightweight Platform for Energy Prediction on Low-Power Edge Devices. Electronics, 14(10), 1912. https://doi.org/10.3390/electronics14101912

Jebur, S. A., Hussein, K. A., Hoomod, H. K., Alzubaidi, L., & Santamaría, J. (2023). Review on Deep Learning Approaches for Anomaly Event Detection in Video Surveillance. Electronics, 12(1), 29. https://doi.org/10.3390/electronics12010029