234 / 2023-09-29 10:15:42

Research on GNSS/INS Vehicle Collaborative Localization Method Based on Variational Bayesian UKF

Collaborative localization,,Integrated Navigation,UKF,variational bayes

  With the development of the navigation and localization field, multi-sensor fusion has been widely used, and the consequent emergence of the cooperative positioning algorithm of vehicle ad hoc networks（VANETs）is a common approach. Global Navigation Satellite System (GNSS), as a mature and widely applicable navigation system, has become the first choice, however, in urban environments, the blockage of GNSS satellite signals reduces the accuracy of positioning results. Inertial Navigation System (INS), as a positioning method that is not affected by external factors, does not rely on any external information and can provide multiple positioning data with high accuracy in the short term. However, long-term operation can cause serious drift in the positioning results. Therefore, GNSS and INS have good complementarity, and the combination of these two systems becomes recognized as an optimal solution. Collaborative localization through fusion of INS and GNSS pseudoranges and Doppler shifts data of participating vehicles.

  In the cooperative positioning based on GNSS/INS, the positioning problem is solved by combining multiple systems. However, the system composed in this way has highly nonlinear characteristics, and the interference error between systems and the measurement noise generated when sensors collect data in a complex and changing unfamiliar environment leads to serious degradation of positioning accuracy. For the nonlinearity of the system, the unscented Kalman filtering (UKF) is used for filtering correction. As for the measurement noise in the filtering system, the traditional method assumes the noise as a uniformly distributed Gaussian noise and applies it to Kalman filtering for carrier position estimation. However, this method cannot regulate the measurement noise anomaly, which affects the positioning accuracy. Aiming at the above problems, this paper proposes a co-localization algorithm for UKF based on variational Bayes. In the process of filtering and updating, the changing observation noise statistics are used as random variables together with the unknown measurement noise, and the probability density function of the system measurement noise variance matrix is estimated and modeled by the inverse Wishart distribution with a reconciled mean using the variational Bayesian algorithm, which alters the Gaussian characteristics of the measurement noise so that it better reflects the real noise. Finally, the algorithm proposed in this article was validated through vehicle collaborative experiments. The experimental results showed that the proposed algorithm can effectively improve its accuracy and robustness compared to the traditional UKF based GNSS/INS collaborative positioning algorithm and the collaborative positioning algorithm without INS.