Dynamic 3D orientation based on constrained extended Kalman filtering using a MARG sensor

Abstract

Although current data fusion techniques for estimating 3D orientation using a magnetic, angular rate, and gravity (MARG) sensor show high accuracy under static motion, the accuracy is deteriorated under dynamic motion due to the external acceleration and change of magnetic dip angle. This study proposes a 3D orientation estimation method based on constrained extended Kalman filtering to overcome the external acceleration and varying magnetic dip angle problems under dynamic motion. The proposed method estimates the external acceleration and change of magnetic dip angle and constrains the norm of the DCM to unit vector to estimate directional cosine matrix (DCM). A series of lab scale experiments were conducted to verify the accuracy and the results were compared to other approaches: (1) a conventional extended Kalman filter, (2) a complimentary filter based on a gradient descent algorithm (Madgwick algorithm), (3) an extended Kalman filter based on a switching approach (4) an extended Kalman filter based on an acceleration model approach. While the proposed method had the accuracy similar to other approaches under static motion, it showed better performance under dynamic motion. Keywords: 3D orientation, Constrained extended Kalman filter, MARG sensor, Data fusion