Development of Edge Computing in CCTV Cameras for Intelligent Video Analytics with a Privacy-Preserving Approach

Development of Edge Computing in CCTV Cameras for Intelligent Video Analytics with a Privacy-Preserving Approach

Hadi Mozafari Mohammad Abadi1 Mostafa Abdollahian Dehkordi2

1) Computer Engineering Student, Department of Computer Engineering, National Skills University, Tehran, Iran
2) Department of Electrical and Computer Engineering, National Skills University, Tehran, Iran

IOI Code : AAHD-ABHEBA
Publication : National Conference on Information Technology, Nanotechnology, Artificial Intelligence and Technological Futures Studies(itnaf.ir)
Abstract :
    The proliferation of CCTV cameras across smart cities, healthcare facilities, industrial environments, public transportation systems, educational centers, and critical infrastructures has fundamentally transformed the nature of video surveillance. Cameras that were historically utilized primarily for recording and reviewing incidents have now evolved into active components in the process of detection, analysis, and alerting, empowered by machine learning, deep learning, and edge computing. However, the conventional architecture—reliant on the continuous transmission of raw video to central servers or the cloud—introduces significant challenges, including processing latency, high bandwidth consumption, substantial storage costs, an expanded cyberattack surface, and the violation of individuals privacy. ​Adopting an analytical-developmental approach, this paper investigates the role of edge computing in smart CCTV cameras with a robust emphasis on privacy preservation. The core premise of this study is that the next generation of video surveillance systems must not only see more accurately but must also see more responsibly. In other words, incident detection must be executed with the minimal processing of identifiable data. According to recent studies, privacy preservation in video analytics is not merely limited to face blurring; rather, it must be addressed across multiple layers, including raw data, extracted features, machine learning models, network communications, and human access policies. ​This paper first examines Cloud, Edge, and Cloud-Edge-Terminal architectures, followed by an analysis of methods such as real-time anonymization, non-reconstructible feature extraction, federated learning, and differential privacy. Furthermore, a proposed framework entitled the Event-Driven Privacy Twin is introduced. Within this framework, instead of storing and transmitting full images of individuals, the system generates an abstract, minimal, and actionable representation of each event—a representation that is sufficient for operational response yet maximally insufficient for direct personal identification. This approach holds significant application potential in environments such as smart hospitals, schools, elderly care centers, subways, airports, critical industrial facilities, and smart cities.

Keywords : edge computing cctv cameras intelligent video analytics privacy preservation edge ai