Home // CYBER 2025, The Tenth International Conference on Cyber-Technologies and Cyber-Systems // View article

A Privacy-preserving Video Processing Pipeline

Authors:
Gábor György Gulyás
Gergely Erdődi

Keywords: cctv; video processing; face recognition; architecture.

Abstract:
We present a modular and scalable video analytics system designed for object detection, face recognition, and multi-camera tracking, with minimal bandwidth consumption and full compatibility with existing video surveillance infrastructure. The architecture emphasizes cost efficiency and regulatory compliance, operating primarily on on-premise deployments to align with constraints imposed by the EU AI Act and GDPR. After benchmarking a range of object detection, face analysis, and tracking models, we selected the most performant and efficient solutions and orchestrated them using Apache Airflow. The system executes a graph-based processing pipeline that supports parallel, per-camera analytics including people counting, path tracking, heatmap generation, and geofencing. Results are visualized through Apache Superset dashboards, enabling interactive, building-wide situational awareness. By leveraging open source components and a containerized, Kubernetes-compatible deployment model, the solution provides real-time, bandwidth-aware analytics with strong adaptability to diverse operational environments, supporting data-driven decision-making across sectors, such as retail, logistics, and smart infrastructure.

Pages: 50 to 54

Copyright: Copyright (c) IARIA, 2025

Publication date: September 28, 2025

Published in: conference

ISSN: 2519-8599

ISBN: 978-1-68558-295-1

Location: Lisbon, Portugal

Dates: from September 28, 2025 to October 2, 2025