Privacy-Preserving Edge AI Security Architecture for Consumer Electronics in Cultural Heritage Environments

The growing reliance on consumer electronics has created new opportunities for affordable and intelligent security solutions, yet balancing effective surveillance with strict privacy requirements remains a critical challenge. Traditional systems, such as CCTV and enterprise-grade thermal cameras, achieve high detection accuracy but often compromise visitor privacy and are cost-prohibitive. Meanwhile, IoT-based alternatives frequently depend on cloud infrastructure or specialized hardware that is unsuitable for resource-limited institutions. To address these challenges, in this paper, we present Privacy-Preserving Edge AI Security for Cultural Heritage using Consumer Electronics (PREIS-CE), a privacy-preserving edge AI architecture that achieves enterprise-level security using only e250 consumer devices. Our system integrates Android TV boxes, WiFi 6 mesh routers, and low-cost thermal and radar sensors through a multi-modal fusion pipeline designed for the limitations of consumer processors. Lightweight neural models are co-designed with formal privacy mechanisms, including R´enyi differential privacy and topology-aware secure aggregation, enabling real-time inference and rigorous protection of personal data. We validate PREIS-CE across three museum deployments, demonstrating over 90% F1-score across five security-critical tasks, formal (ε = 1.83, δ = 10−5)-differential privacy guarantees, 47ms latency, 5.8W power consumption, and reliable operation over 2.4 million processed events.