Adversarially-robust multi-biometric fusion with semantic-aware decision making for home intruder detection
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Biometric systems used in security-sensitive environments require not only high recognition accuracy but also robustness to adversarial manipulation and reliable, privacy-preserving data handling. This work presents the Multi-Biometric Intruder Detection and Home Safety System (MBID-HSS), which integrates gait, face, and iris recognition through modality-specific encoders and a supervised contrastive fusion module. A Transformer-based classifier performs identity prediction, with a lightweight alternative available for latency-constrained deployments. Optional semantic-reasoning and blockchain-backed logging components enable context-aware decision support and auditable event recording without storing raw biometric data. Evaluation is performed on a composite benchmark constructed from seven public datasets, where each modality constitutes an independent identity space and subject-disjoint splits are strictly enforced. Under a unified and fully reproducible training protocol, multimodal fusion consistently outperforms single-modality baselines, achieving 99.2% accuracy. Adversarial robustness is quantified using standard gradient-based attack models, and system latency is measured on embedded hardware to assess deployment feasibility. Overall, MBID-HSS provides a transparent and reproducible system-level integration of established biometric recognition, adversarial-defense, and contextual-reasoning techniques for practical multi-biometric home-security applications.
Fields of Study & Subjects
Biometric Identification and Security
Gait Recognition and Analysis
Face recognition and analysis
Publication details
- Journal / source
- Egyptian Informatics Journal
- Volume
- 34
- Pages
- 100942–100942
- Publication date
- 27 March 2026
- Language
- EN