With the rapid growth of smart home technologies and intelligent surveillance systems, ensuring security while protecting user privacy has become increasingly important.Traditional surveillance systems usually rely on centralized servers where video data is continuously stored and processed.However, this method can expose sensitive personal information, increase communication costs, and create risks related to data leakage and unauthorized access.To overcome these challenges, the proposed work introduces a hybrid software-based and hardware-assisted Home Intrusion Detection System that uses Federated Learning for secure and efficient surveillance.In this system, multiple smart home environments are simulated as virtual clients within a single platform.Each client captures video data from cameras and trains its own local intrusion detection model independently.Instead of sending raw video footage to a central server, only model parameters are shared, helping preserve user privacy and reduce network overhead.The framework applies computer vision and anomaly detection techniques to identify suspicious activities such as unauthorized access, unknown persons, and unusual movements.A central server aggregates the locally trained models to create a global model, which is redistributed to clients for continuous improvement.Overall, the system provides a scalable, privacy-preserving, and cost-effective solution for next-generation smart home security.
Kavitha et al. (Thu,) studied this question.