A Direct Vehicle Tracking Algorithm Based on Adaptive Parallel Factor Decomposition

Vehicle positioning and tracking play an essential role in intelligent transportation systems, especially under the growing demands of autonomous driving, traffic management, and path planning. However, most existing works adopt classical two-step methods and suffer from suboptimal performance due to intermediate estimation errors. In this work, we propose a direct vehicle tracking method based on adaptive PARAllel FACtor (PARAFAC) decomposition, referred to as DT-AP, which differs from conventional tracking frameworks by operating directly on the received signal. We first reveal that the received signals can be naturally modeled as a low-rank dynamic streaming tensor, representing the multi-dimensional and time-evolving characteristics of vehicle motion in distributed sensing systems supported by 5G ultra-dense networks. By adaptively decomposing the streaming tensor, DT-AP enables online trajectory estimation while eliminating intermediate processing and thereby reducing information loss. Numerical simulations are conducted to validate the effectiveness of the proposed method under dynamic multipath environments. Simulating results demonstrate that DT-AP outperforms traditional tracking approaches in both accuracy and computational complexity, while maintaining robustness under multipath conditions. These features indicate its potential for real-time and reliable applications in intelligent transportation systems.