Adaptive EKF Based on HMM Recognizer for Attitude Estimation Using MEMS MARG Sensors

Abstract

This paper addresses the intractable problem of attaining precise attitude estimation efficiently using MEMS MARG (magnetic, angular rate, and gravity) sensors for 3-D motion tracking, which is called attitude and heading reference system (AHRS). The performance of AHRS is adversely affected by sensor noise and measurement disturbances. To overcome this problem, we propose a novel adaptive extended Kalman filter (AEKF) in this paper, including a multiplicative extended Kalman filter (MEKF) and a hidden Markov Model (HMM) recognizer. MEKF is built on sensor noise model. An HMM recognizer is developed to identify the measurement disturbance caused by motion or environmental interference and then to adjust the noise covariance of MEKF adaptively. The proposed AEKF was assessed with a high-precision test platform by using a high-caliber commercial AHRS. Experimental results indicate that the proposed method achieves a higher level of stability and accuracy in comparison to other attitude estimators, especially in dynamic tests, which demonstrates the availability of 3-D motion tracking.