Improving the accuracy of distributed cooperative positioning at urban intersections based on mathematical analysis and robust filtering

To improve the cooperative positioning accuracy of urban intersections under conditions of global positioning system signal blockage and complex traffic flow, this paper proposes a cooperative positioning model that integrates a mathematical analytical Gaussian belief propagation robust filtering algorithm with a spatial attention mechanism. A hardware verification platform integrating roadside sensing units, connected test vehicles, and 5G-vehicle-to-everything communication is constructed to achieve distributed cooperative positioning analysis of urban intersections. The results show that the proposed fusion model exhibits the best overall performance in intersection cooperative positioning, with a root mean square error of only 0.15 m, significantly lower than both single algorithms and the spatiotemporal attention model. The geometric accuracy attenuation factor is as low as 1.8, the effective cooperative gain reaches 60%, and the positioning coverage rate reaches 98%. However, experimental results under ring-topology, high-density congestion, and communication failure scenarios indicate a noticeable performance degradation, suggesting that the robustness of the proposed method is limited under extreme network conditions. The integrated design of “mathematical model analysis - dynamic feature extraction - vehicle-road cooperative fusion” can effectively achieve a leap in intersection positioning accuracy from single-vehicle intelligence to group collaboration, providing key technical support for reliable perception of smart intersections and connected autonomous driving.