Adaptive Expectation–Maximization-Based Kalman Filter/Finite Impulse Response Filter for MEMS-INS-Based Posture Capture of Human Upper Limbs

Abstract

To obtain precise positional information, in this study, we propose an adaptive expectation–maximization (EM)-based Kalman filter (KF)/finite impulse response (FIR) integrated filter for inertial navigation system (INS)-based posture capture of human upper limbs. Initially, a data fusion model for wrist and elbow position is developed. Subsequently, the Mahalanobis distance is utilized to evaluate the performance of the filter. The integrated filter employs the EM-based KF to enhance noise estimation accuracy when the performance of KF declines. Conversely, upon deterioration in the performance of the EM-based KF, which is evaluated using the Mahalanobis distance, the FIR filter is employed to maintain the effectiveness of the data fusion filter. This research utilizes the proposed EM-based KF/FIR integrated filter to ascertain wrist and elbow positions. The empirical results demonstrate the proficiency of the proposed approach in estimating these positions, thereby overcoming the challenge and highlighting its inherent effectiveness.