Fusion GNSS/INS/Vision With Path Planning Prior for High Precision Navigation in Complex Environment

With the rapid advancement of autonomous driving technology, achieving accurate and continuous navigation in complex urban environments has become a critical research area. However, challenges such as dynamic objects and complex lighting conditions can significantly compromise the accuracy of visual-inertial odometry. In addition, structures such as tunnels, overpasses, and urban canyons may disrupt the continuity and accuracy of global navigation satellite system (GNSS) signals. This article presents a tightly coupled model that integrates GNSS, inertial navigation system (INS), vision observations, and lightweight map priors. This system leverages prior trajectory information from the maps to generate discrete control points, ensuring adherence to designated roads. It also utilizes deep learning algorithms to identify and exclude dynamic objects, thus enhancing the reliability of visual odometry calculations. This approach maintains decimeter-level accuracy even during prolonged GNSS signal outages and facilitates rapid convergence of precise point positioning (PPP). In urban scenarios featuring one long tunnel, our algorithm demonstrated robustness with an overall root mean square error (RMSE) of 0.9 m, showcasing its effectiveness in challenging conditions with frequent GNSS disruptions. In environments without tunnels, the accuracy is further improved, achieving an RMSE of 10 cm, which surpasses the accuracy of standalone PPP solutions.