HYBRID ARCHITECTURE WITH ADAPTIVE LATENCY FOR REAL-TIME VIDEO SURVEILLANCE ANALYSIS INTEGRATED WITH SOCIAL NETWORK DATA

Mabouh Moise Hermann

The integration of video surveillance systems with social network analysis presents a significant challenge for modern computing architecture due to conflicting requirements for latency, accuracy, and scalability. Existing approaches consider these domains separately, creating a technological gap that limits the effectiveness of integrated security systems. We propose hybrid architecture with adaptive latency that strategically combines the advantages of microservice and edge-cloud architecture. Our methodology includes the implementation of an adaptive model partitioning system, a context-based prioritized processing pipeline, and an asynchronous fusion mechanism with consistency guarantees. Tests were conducted on three configurations of increasing scale (50, 250, and 1,000 cameras). The proposed architecture demonstrates a 43% reduction in overall latency compared to conventional architectures, a 38% improvement in resource efficiency, and a 7.2% increase in cross-identification accuracy. Composite efficiency, combining these metrics, achieves a 52% improvement for the large-scale configuration. The hybrid approach with adaptive latency represents significant progress for integrated surveillance systems. Its capability for contextual adaptation and resource efficiency makes it particularly suitable for existing urban deployments and infrastructures with variable bandwidth constraints, offering a realistic path to improve current systems without completely redesign.

Keywords:video surveillance, social network analysis, hybrid architecture, distributed processing, adaptive latency, model partitioning, multimodal fusion, real-time systems, integrated security, edge computing