An efficient bias estimation method in multisensor fusion for navigation by adaptive prototype selection in a bank of Kalman filters

Abstract

A navigation system usually employs multiple sensors to combine the strength of individual sensors such as GPS, gyroscope, and accelerometer. However, the multiple sensor fusion for navigation objectives encounters noise and time-variant bias present in individual sensor measurement. This paper proposes an efficient method to estimate the unknown bias for cancellation in a fused navigation system involving multiple sensors using an adaptive bias prototype selection employed over a bank of parallel Kalman filters. The conventional method which focuses on a special case of bias characteristics revolving around the semi-Markov process model is recognized for its excessive computation if the bias prototype set is chosen too large. This means that a huge discrete set of bias is needed to obtain the bias estimation accurately. Focusing on solving the problem of large bias prototypes, we propose a two-step selection process: (1) the decision part that locks on a new bias set from estimated bias covariance and (2) the balance part that regulates the newly selected bias set to enable a smooth transition under inadvertent bias overshoots. Simulation results show a substantial improvement in bias estimation accuracy while maintaining a minimal computation compared to the nonadaptive randomly switching bias estimators.