Initializing An Unscented Kalman Filter Using A Particle Filter

This work develops an algorithm to initialize an Unscented Kalman Filter using a Particle Filter for applications with initial non-Gaussian probability density functions. The method is applied to estimating the position of a road vehicle along a one-mile test track using terrain-based localization where the pitch response of the vehicle is compared to a premeasured pitch map of the test track. The results indicate that the method can be used to decrease the computational load of the algorithm while maintaining the accuracy of the Particle Filter. A modified Chi-Squared test is optimized and used to determine a switchover point when the probability density function of the particle population can be approximated by a Gaussian for initializing the Unscented Kalman Filter.