Digital twin for city traffic emissions using deep vision and real time monitoring

Urban on-road air pollution remains a pervasive threat to public health, yet its management is fundamentally constrained by a lack of spatiotemporal precision in emission inventories. Conventional inventories rely on static, registration-based proxies that fail to capture the high-frequency fluctuations of traffic-derived pollutants. Here, we present an AI-driven, high-resolution digital twin framework that reconstructs hourly, road-segment-level emissions in Atlanta’s urban core as an example by integrating deep-vision vehicle classification with real-time roadside camera networks. Our findings reveal that traditional inventories systematically underestimate localized air pollution burdens because they neglect the significant impact of inter-city freight transportation. Validated through Monte Carlo uncertainty analysis, this system enables a transition from retrospective accounting to real-time operational monitoring. By pinpointing corridor-specific hotspots and resolving the complexities of fleet-driven pollutant burdens, this research provides a scalable architecture for hyper-local exposure forecasting and the targeted mitigation of environmental health disparities in global cities.