Deep Trajectory Recovery Approach of Offline Vehicles in the Internet of Vehicles

Xiao Han, Ding-Xuan Zhou, Guojiang Shen, Xiangjie Kong, Yulong Zhao

Abstract: Multi-modal trajectory analysis and trajectory recovery are essential tasks in transportation research, especially for offline vehicles, which enable comprehensive understanding of complex transportation systems and address the issue of incomplete or missing trajectory data. In this paper, we propose a novel Deep Trajectory Recovery Framework, DTRF, which can effectively tackle both challenges by using a combination of a Cellular Automata (CA) model and a Multi-Kernel Graph Neural Network (MKGNN) model. The CA model plays a crucial role in normalizing and representing multi-modal traffic data with diverse structures, sampling frequencies, and physical meanings. By capturing the inherent relationships among different modalities, the CA model enables our proposed framework to make better use of these multi-modal data from networked vehicles and roadside detectors and then generate data for traditional vehicles. The MKGNN model, built on the foundation of spectral graph theory, employs various kernels to model different driving characteristics. The use of multiple kernels allows the GNN model to capture a wide range of driving patterns, enhancing its ability to reconstruct missing trajectories accurately. To validate the effectiveness of our proposed model, extensive experiments are conducted on two datasets. The results demonstrate that our framework outperforms state-of-the-art baselines in terms of trajectory recovery, showcasing its efficiency and robustness.