A GNSS/INS Robust Positioning Method Based on GML Mixture Distribution and Improved VB in Complex Urban Environment

The global navigation satellite system (GNSS) and inertial navigation system (INS) achieve precise, real-time and reliable positioning through information fusion, which is the fundamental and key technology for new-generation information technologies such as driverless vehicles, intelligent transportation and smart cities. However, in complex urban environment, GNSS signals are prone to generating outlier observations, and these outlier observations occur randomly. The GNSS observation noise presents non-stationary characteristics, resulting in a significant decline in the positioning performance of the GNSS/INS integrated system. A GNSS/INS robust positioning method based on the Gaussian-multivariate Laplacian (GML) mixture distribution and the improved variational Bayesian (VB) is proposed. Firstly, in view of the unpredictable generation characteristics of GNSS outlier observations, a GNSS observation noise model based on the GML mixture distribution is constructed, and a mixture parameters determination strategy based on the innovation vector is designed; Then, aiming at the problem that the traditional VB has high computational complexity in solving the relevant parameters of the GNSS observation noise model and the posterior state vector, an improved VB is proposed to reduce the computational load and improve the real-time positioning performance of the integrated system; Finally, the results of the vehicle-mounted road experiment show that, compared with the existing methods, the positioning accuracy of the proposed method in the slight urban canyon environment has increased by 16.03%, and in the severe urban canyon environment, it has increased by 21.94%. The computational complexity of the improved VB is reduced by 63.68% compared with the traditional VB. This method is also applicable to other information fusion fields that are disturbed by outliers.