RSUTrajRec: Multi-granularity trajectory recovery based on roadside units sensing

Abstract

Vehicle mobility trajectories, especially fine-grained trajectories, provide valuable insights for understanding urban dynamics and play a crucial role in intelligent transportation systems and urban planning. Obtaining fine-grained vehicle trajectories can be realized by trajectory recovery, but traditional efforts suffer from defects such as poor privacy protection and low recovery accuracy. To address these issues, we propose a new scenario of trajectory recovery based on roadside unit (RSU) sensing. However, this scenario introduces a significant challenge: recovering high-precision trajectories from the incomplete and unevenly distributed sensing data. To tackle this, we design RSUTrajRec, a multi-granularity trajectory recovery framework that comprises a graph neural network-based module for road information prediction, a Transformer-based module for multi-granularity recovery, and an RSU deployment planning module. Extensive real-world dataset evaluations reveal that RSUTrajRec has a significant advantage in recovering missing vehicle trajectories outside the RSU coverage area. In addition, evaluations also verify that the performance of the trajectory recovery task can be effectively improved by optimizing the RSU deployment plan.