On Filtering Algorithm Of UKF Based GPS/DR Integrated Vehicle Navigation System

With increasing of vehicles, the traffic jam, energy waste and traffic accident have become seriously. Vehicle navigation system can solve these problems efficiently. GPS and Dead Reckoning system are the most common used vehicle navigation systems. A GPS/DR integration provides position information with high reliability for vehicle navigation system.In order to get high accuracy position information, it is very necessary to research the data-filtering method. Kalman Filtering is the most successful data-filtering method to deal with the problem of multi-sensor data fusion. Extended Kalman Filtering and Unscented Kalman Filtering are the most common used nonlinear filtering methods.Firstly, the characteristics of Extended Kalman Filtering and Unscented Kalman Filtering have been analyzed. Extended Kalman Filtering is a traditional nonlinear filtering method. It needs to linearize the nonlinear model and obtain the Jacobian matrix, but sometimes Jacobin matrix is not available for some systems. While Unscented Kalman Filtering doesn't need to linearize the nonlinear system model and doesn't need to get the Jacobian matrix. And Unscented Kalman Filtering can get more accuracy estimate value and has similar computation complexity with EKF.Secondly, Federated Kalman Filtering is applied to GPS/DR integrated navigation system. It filters the GPS and DR system respectively. This kind of Kalman Filtering can reduce the computation. The mathematical model of GPS is linear, but the DR system is nonlinear. In the GPS sub-filter, traditional linear Kalman Filter is used, while nonlinear filer is used in the DR system.Thirdly, five sampling strategies are introduced in the thesis. They are respectively based on skewed simplex sampling strategy, spherical simplex sampling strategy, symmetry sampling strategy, extended sampling strategy, and fourth order moment strategy. The differences are investigated between the five sampling strategies from the estimate accuracy and computation complexity.Finally, matlab is used to simulate the GPS/DR integrated navigation system based on Extended Kalman Filtering and Unscented Kalman Filtering, and analyze the five different sampling strategies. The estimated value is compared with measured value and true value of the six filtering methods from eastern velocity and position.