Tilt Compensation of Inertial Sensors With Particle Flow Filters’ Fusion

Inertial Measurement Units (IMU) are commonly used in many applications. However, these sensors experience significant deterioration in the orientation angles when (tilting) moving from their stationary state on the ground, containing vertical axis components. The algorithm that addresses this issue is called “tilt compensation of inertial sensor/IMU”. The prevailing methods, which are mostly performed by projecting the measurements onto a horizontal plane, have some limitations, notably for relatively large tilt angles. On the other hand, non-linear filtering, especially for high-dimensional systems` state estimations, is still a significant problem. In the last decade, the “Flow of Particles” has provided a novel approach to the issue of solving the dimensionality of the standard particle filter for state estimation in nonlinear systems. Nevertheless, Particle Flow Filter (PFF) employment for IMU tilt compensation has not yet been seen in the literature. Therefore, for tilt compensation of inertial measurement-sensor systems, the Particle Flow Filter structure here is used for the first time in the literature by this study. The approach for AHRS (Attitude and Heading Reference Systems) in which acceleration, rotation, and magnetometer data are fused includes probabilistic filtering, and tilt compensation algorithm, alongside a magnetometer calibration routine, that determines factors in orientation accuracy. The simulation results show that the algorithm made a more significant orientation accuracy improvement, which was shown in detail by the metrics, than others in vertical movements of less than 90 degrees for test-raw sensor data. Performance differences between the method and previous current approaches are highlighted here as well. The method here outperforms the conventional tilt-compensation techniques, particularly Kalman-based complementary filters (CF) and modifiers such as Madgwick, in terms of yaw estimation accuracy, as evidenced by less than 2° RMS- the root mean square error. Although the Particle Flow Filter-based algorithm has some computational complexity with more processing time, the results here make it clear that it outperforms others in terms of tilt-compensation orientation accuracy. Further studies with more robust and novel PFF versions’ filter structures can achieve high accuracy and precision performance even with unreliable low-cost sensors.