Kalman filter based integration of DGPS and vehicle sensors for localization

Real time implementation of a Kalman filter based estimator is presented. Differential GPS data is fused with vehicle sensors's data to achieve decimeter-level accuracy position and degree-level heading estimation for the car. Vehicles sensors consist of wheel speed, steering angle and yaw rate. A dynamic bicycle model is utilized as the process model. Due to its nonlinearities we use an extended Kalman filter. GPS noise is not white Gaussian in practice. Different kinds of GPS noise are discussed. Experimental results on the performance of the filter are presented.