Tel2Veh: Fusion of Telecom Data and Vehicle Flow to Predict Camera-Free Traffic via a Spatio-Temporal Framework

Predicting vehicle flow is crucial for traffic management but is often limited by the scope of sensors. In contrast, extensive mobile network coverage enables us to utilize counts of mobile users' network activities (cellular traffic) on roadways as a proxy for vehicle flow. However, cellular traffic counts, which encompass various user types, may not directly align with vehicle flow. To address this issue, we present a new task: utilizing cellular traffic to predict vehicle flow in camera-free areas. This is supported by our Tel2Veh dataset, which comprises extensive cellular traffic and sparse vehicle flows. To tackle this task, we propose a two-stage framework. It first independently extracts features from multimodal data, and then integrates them using a graph neural network (GNN)-based fusion to generate predictions of vehicle flow in camera-free areas. We pioneer the fusion of telecom and vision-based data, paving the way for future expansions in web-based applications and systems.