Low-cost PPP/visual/IMU fused positioning method based on factor graph optimization under complex urban environments

With the development and application of autonomous driving technology, high-precision positioning and navigation services have played an increasingly important role in complex urban environments. The visual/inertial combination algorithm can utilize the spatial structure and texture information of the environment to achieve robust pose estimation in complex urban environments. However, it suffers from cumulative errors in large-scale scenes and an inability to obtain global positioning information. In the global navigation satellite system (GNSS), precise point positioning (PPP) can achieve centimeter-level accuracy in open environments, but in complex urban environments, degraded accuracy or failure may frequently occur due to satellite signal obstruction. To leverage the complementary strengths of GNSS and visual-inertial odometry (VIO) to achieve high-precision positioning in complex urban environments, this paper proposes a PPP/visual/IMU combination algorithm. Firstly, the algorithm uses the joint initialization of multi-sensor to achieve the spatial-temporal alignment. Then, a PPP position error factor was constructed to achieve joint backend optimization of PPP, visual, and inertial. Furthermore, apply adaptive weighting based on the number of visible satellites, position dilution of precision (PDOP) value and variance of PPP observations to ensure that the system can perform continuous and high-precision positioning in environments where GNSS signals are obstructed or degraded. Finally, accurate and robust estimation of global pose was achieved through factor graph optimization based on keyframes and sliding windows. We use a public dataset to verify the usability of our work. Real-world experiments show that, in an open environment, the proposed PPP/visual/IMU combination system has comparable accuracy to PPP. In a typical urban complex environment, our system can improve accuracy by 11.35% compared to PPP. In the GNSS degradation environment, our system can improve accuracy by 19.65% compared to PPP.