RTK-EVC: Real-Time Kinematic Positioning With Enhanced Velocity Constraints Via Factor Graph Optimization in GNSS-Challenged Environments

In this paper, we propose a novel Real-Time Kinematic positioning architecture with Enhanced Velocity Constraints (RTK-EVC) which is solved by factor graph optimization. The RTK procedure has been redesigned, which consists of three phases: the velocity estimation, the position estimation and the Partial Ambiguity Resolution (PAR) augmented by the Dead Reckoning (DR) consistency check. The velocity estimation, which provides high-accuracy velocity information to construct velocity constraints, is connected with the position estimation sequentially when they are used for calculating the float solution. The DR consistency check is proposed to enhance the reliability of the PAR. A dual-part factor graph architecture is designed for FGO-RTK. The whole factor graph is divided to the velocity factor graph and position factor graph which are solved sequentially to prepare velocities for velocity constraints in position estimation. In the velocity estimation, RANdom SAmple Consensus (RANSAC) Doppler-shift quality control constant velocity hypothesis factors and constant acceleration hypothesis factors are implemented to formulate a robust and accurate velocity estimator. In the position estimation, velocity-independent constraint factors are employed for decoupling the estimation of velocity from position. Fixed prior factors are designed to maintain the high precision information of the historical fixed epochs. As for the PAR augmented by the DR consistency check, the fixed solution, which has passed the conventional ratio test, is further checked by the DR results from the historical fixed epochs. The new architecture is evaluated by both open-source dataset and fresh experimental dataset. It is demonstrated that the RTK-EVC architecture can outperform current open-source methods in GNSS-challenged environments. Different strategies in the RTK-EVC architecture are also comparatively analyzed. The C++ implementation of the RTK-EVC and the new dataset can be gotten via https://github.com/zhaoqj23/RTK-EVC.