@article{Liu2025, 
author = {Yanbin Liu and Bolin Gao and Peikun Lin and Guangyu Tian and Keqiang Li},
title = {A framework for real-time vehicle tracking in large-scale roadside sensor networks},
year = {2025},
journal = {Green Energy and Intelligent Transportation},
volume = {4},
number = {6},
keywords = {Vehicle-Road-Cloud Integration System (VRCIS), Vehicle Trajectory, Multiple-Target Tracking, Multi-Access Edge Computing (MEC)},
url = {https://www.sciopen.com/article/10.1016/j.geits.2025.100362},
doi = {10.1016/j.geits.2025.100362},
abstract = {Vehicle-Road-Cloud Integration system (VRCIS) requires high-precision vehicle positioning and tracking, with very low system latency, which is a difficult task given the quantity and quality of data. To this end, a distributed computing framework using multi-access edge computing (MEC) devices is proposed in this paper. To process trajectory data (including preprocessing, calibration, multi-sensor trajectory matching, and trajectory prediction), as well as integrate machine learning algorithms to improve the accuracy of trajectory prediction, especially for complex and diverse driving scenarios environmental conditions, a framework is designed. In addition, to conduct a comprehensive evaluation of the overall performance of trajectory tracking, factors such as trajectory smoothness and velocity consistency — components of our novel evaluation metrics — are considered. Experiments show that the framework can continuously track tens of thousands of vehicles on highway, with average longitudinal and lateral errors of 2.14 and 0.84 ​m respectively, with average speed error of 1.91 kph. The experiments on large-scale road networks with 1,777 sensors are implemented, with continuous multi-vehicle tracking over 157 ​km of highway, and establishing superior performance compared to existing methods. Furthermore, processing latency remained below 340 ​ms, demonstrating the potential of this framework to enhance driver experience, improve road safety and efficiency.}
}