ITERATIVE KALMAN FILTERING

摘要

Several types of noise limit the performance of remote sensing systems, e.g., systems that determine the location, color, or shape of remote objects. When noise detected by sensors of the remote sensing systems is known and well estimated, a Kalman filter can converge on an accurate value without noise. However, non-Gaussian noise bursts can cause the Kalman filter to diverge from an accurate value. Current approaches arbitrarily boost noise with fixed additive or multiplicative factors. Such approaches slow filter response and; thus, often fail to give timely results. Such noise boosts prevent divergence due to badly corrupted measurements. Disclosed embodiments eliminate a subset of noise measurements having the largest errors from a data set of noise measurements and process the remaining data through the Kalman filter. Advantageously, disclosed embodiments enable a Kalman filter to converge on an accurate value without the introduction of noise boost estimates, which adds processing time.

主权项

1. A method for improving a Kalman filter solution, comprising:
(a) receiving a plurality of sequential measurements of a state variable, each measurement including a respective noise contribution; (b) obtaining for each measurement of the plurality of sequential measurements, a Kalman-filtered estimate of the state variable; (c) determining for each Kalman-filtered estimate, a respective error indicative of a departure of the Kalman-filtered estimate from the state variable; (d) selectively removing at least one sequential measurement from the plurality of sequential measurements responsive to a predetermined condition imposed upon the respective errors, those sequential measurements remaining after removal being referred to as a tailored plurality of sequential measurements; and (e) obtaining for each measurement of the tailored plurality of sequential measurements, a Kalman-filtered estimate for the state variable.