Posture estimation system by IMM-based unscented Kalman filters

摘要

A real-time posture estimation system by interacting multiple model (IMM) based unscented Kalman filters (UKFs) is presented in this paper. The system is developed on an inertial measurement unit (IMU) development board which is integrated with multi-sensors, microcontroller, and WIFI module, aiming at obtaining high accuracy posture estimation. In order to reduce the overall estimation error of the system, the motion data of the sensors are processed by wavelet denoising method firstly. Then, the IMM-based unscented Kalman filters including quaternion-based unscented Kalman filter (QBUKF) and gyroscope-based unscented Kalman filter (GBUKF) is adopted. The IMM algorithm is capable of adaptively fusing the estimated posture from the two models, and generating the optimal estimation via the flexibility of model weighting. The optimal estimation is transmitted to the host computer through WIFI, which effectively solves the inconvenience of wired transmission. System performance is evaluated by experimental tests using the robot UR5 as the truth reference system. The results demonstrate the feasibility of the hardware scheme and the effectiveness of the proposed algorithm.