A systematic review of AI powered adaptive smart home security using multimodal sensor fusion edge intelligence and privacy preserving architectures

Abstract

The rapid proliferation of IoT devices and AI-driven sensing has transformed modern residences into intelligent cyber-physical environments, simultaneously expanding their attack surface to physical intrusions, cyberattacks, sensor manipulation, and privacy violations. Conventional single-modality, rule-based security systems are increasingly inadequate for such dynamic environments. This paper presents a systematic review of AI-powered adaptive smart home security systems based on multimodal sensor fusion, synthesizing 87 primary studies identified across IEEE Xplore, Scopus, Web of Science, ScienceDirect, and Google Scholar (2018–2026). The review examines heterogeneous sensing modalities including RGB and thermal cameras, PIR sensors, mmWave radar, WiFi CSI, acoustic sensors, smart locks, wearables, and environmental sensors, and critically analyzes multimodal data engineering, fusion strategies (early, late, hybrid, attention-based), adaptive intelligence, Edge AI deployment, and privacy-preserving architectures. Emerging directions including transformer-based fusion, federated learning, foundation models, and digital twins are also discussed. Key findings reveal that current solutions remain fragmented across sensing, fusion, adaptation, and privacy dimensions, and that future research must prioritize unified frameworks jointly addressing multimodal fusion, continual adaptation, edge inference, privacy preservation, and explainability.