Efficient Quaternion Attitude Estimation Algorithm for Particle Filter Based Localization

摘要

The presented work develops a high-dynamic quaternion-based Unscented Kalman Filter (UKF), suitable for implementation in a low-performance system comprising of low-cost MEMS inertial and magnetic sensors. Computational issues of the conventional UKF are addressed by employing a reduced spherical sigma-point selection strategy and propagation of square-root covariance matrix. Angular rate is treated as a control input rather than a part of the state, thus preserving potentially high dynamics. Gyroscope bias and translational acceleration are proposed to be included to the estimated state for the case of low-performance system and typical human motion. Additionally an indoor localization system is presented, which utilizes those orientation estimates together with radio based range measurements and an indoor map. A Sequential Monte Carlo based localization algorithm is developed to combine all available information and estimate the position of a person moving through a building.