Regression smoothing kalman filter

Abstract

The results of an experimental comparison of the accuracy of the classic Kalman filter and a simple non-causal smoother are presented, and a new version of the Kalman smoother, which does not need of a time lag, is described. The offered filter is based on a local approximation of a noisy trajectory by parametric curves. The state parameters of the filter are coefficients of the local regression that is used to build a priori and a posteriori estimates. Results of experimental comparison of the new filter with the linear Kalman one are given.