An Optimal Adaptive Kalman Filter with Gain Correction Based on Innovation Outlier Detection for Human Motion Tracking

In this article, an optimal adaptive Kalman filter algorithm is proposed to mitigate the impact of measurement outliers on human motion tracking. The algorithm utilizes an inertial measurement unit (IMU) equipped with a three-axis gyroscope, accelerometer, and magnetometer for tracking human motion. Due to the presence of modeling issues, an adaptive factor is introduced to adjust the process noise covariance matrix, for compensating modeling errors. When the innovation exceeds the specified threshold range, the Kalman gain K is adjusted to compensate for the impact of measurement outliers. Segmented weight functions are constructed to apply appropriate weights to the Kalman gain to suppress the impact of outliers. When the innovation exceeds the threshold range, the gain K is used to correct the posterior covariance and suppress abnormal values. This approach effectively prevents filter divergence that can occur when the continuous innovation sequence is set to zero without adjusting the posterior covariance. The estimated gravity acceleration and geomagnetic field are then used to calculate the Euler angles via a triaxial attitude determination algorithm. Finally, The effectiveness of the proposed method is demonstrated through various action experiments in different motion scenarios, with experimental results showing a significant reduction in errors.