Analyzing the impact of social security systems on video-based public health surveillance

The increasing reliance on automated video-based systems for public health surveillance introduces some significant challenges in environments where social security systems influence health behaviors and outcomes. Motivated by the need to integrate governance structures with health informatics, this study proposes a framework for spatio-temporal health monitoring that explicitly accounts for the interaction between policy measures and population-level behavior. Traditional approaches often struggle to capture the stochastic nature of health-related signals, overlook spatial heterogeneity across communities, and remain insufficiently responsive to evolving policy interventions.

To address these limitations, we develop the hierarchical epidemiological transformer (HET), a deep learning architecture designed to model complex temporal and spatial dependencies in video-derived surveillance data. HET is augmented with a policy-aware dynamic calibration mechanism (PDCM), which incorporates real-time policy signals and statistical deviations to dynamically recalibrate predictions. This framework integrates health indicators, demographic diversity, and policy-driven interventions to support robust anomaly detection and short-term forecasting, while maintaining low-latency inference suitable for real-time deployment.

Empirical evaluations on multiple public health video surveillance datasets spanning different urban regions and policy settings demonstrate that the proposed model achieves consistently strong performance across heterogeneous environments and improves sensitivity to early-stage epidemiological anomalies compared to strong baselines. The approach advances social security-informed health analytics and offers a practical pathway toward more responsive and equitable public health surveillance systems.