Reliable monitoring of structural condition is a fundamental requirement for maintaining the safety and serviceability of civil infrastructure systems. Bayesian filtering techniques are commonly employed in structural health monitoring (SHM) to estimate states and parameters of the system. However, estimating a complete set of parameters often introduces bias and leads to high computational demand. To address these challenges, this work presents a Digital Twin (DT) framework that combines Proper Orthogonal Decomposition (POD) with Bayesian filtering in a unified manner. Damage localization is performed by tracking variations in the slope of the first Proper Orthogonal Mode (POM), which is continuously updated using an online Singular Value Decomposition (SVD) scheme. In parallel, a Bayesian filter is used to estimate system states and update stiffness parameters only at those degrees of freedom (DOFs) identified as damaged, based on deviations between the evolving POM slopes and a reference baseline obtained from the undamaged structure. This selective and coordinated interaction between online SVD and Bayesian filter significantly improves computational efficiency while maintaining estimation accuracy and robustness. The effectiveness of the proposed framework is demonstrated through detailed numerical studies on an eight-story shear building subjected to multiple damage scenarios, including adjacent-story, non-adjacent, and progressive damage cases. The results indicate that the proposed DT-based framework can consistently identify and quantify multiple damage events, providing a scalable and efficient solution for real-time SHM.