Localization Accuracy Analysis for Roadside Sensing System

Localization under GNSS-denied environments has become a significant research focus in recent years. Traditional solutions for accurate positioning often rely on expensive multi-sensor fusion technologies, such as vision, inertial measurement unit (IMU), and lidar, which are primarily used in autonomous vehicles. However, these solutions can be costly and not suitable for all applications. A new paradigm for positioning services is the use of roadside sensing systems (RSS). These systems utilize vision, radar, and/or lidar sensors to detect road users and transmit the information through cellular vehicle-to-everything (C-V2X) technology. Unlike sensor-fusion-based ego-state estimation used in autonomous vehicles, RSS offloads the computation and sensing work to the roadside, providing positioning services to all connected vehicles (CV) at a lower cost. In this work, we focus on the accuracy analysis of a roadside sensing system for localization. We first proposed a spatiotemporal decoupling association method is then used to associate the RSS positioning data with the ground truth, mitigating timing errors. Then the accuracy analysis and model fitting are performed using data collected from multiple RSS deployed in different locations in China. Finally, we establish three different noise models for the general RSS system by performing a 2-D histogram regression on localization errors, and recommend a simple linear model for general RSS. The outcome of this study provides a practical spatial-related confidence model for RSS localization service users, forming a solid foundation for future applications, particularly for the fusion of GNSS and RSS in autonomous and connected vehicle systems.