Selective and Enhanced Privacy-Preserving Surveillance: Real-Time Face Anonymization Using Gaussian Blur and Pixelation

Public surveillance networks are significant for surveillance and security but are hugely privacy-intrusive due to the widespread usage of facial recognition technologies. Traditional privacy-preserving methods of homomorphic encryption and face blurring either compromise on usability or require high computational effort, making them inappropriate for real-time processing. This paper proposes a novel selective anonymization scheme that is both privacy-preserving and functional-integrity-preserving for surveillance videos. Our approach fuses YOLOv8 nano facial detection with facial matching algorithm to determine target and non-target individuals. Non-target faces are anonymized by a combination of pixelation and Gaussian blurring to render full privacy safeguards. Experimental evaluations using the WIDERFACE dataset and Face detection dataset confirm our methodology to achieve 98.7% privacy preservation, 92.3% target recognition accuracy, and process frames on ordinary CPU, making it highly efficient for real-time applications. The work contributes to the development of privacy-aware surveillance systems that are data protection law-compliant without sacrificing security effectiveness.