Abstract: To meet urgent application demands in indoor location and motion tracking, studies on low-cost high-resolution and real-time micro-electro-mechanical system-inertial measurement unit (MEMS-IMU) have attracted much attention. This paper presents a quaternion-based data fusion algorithm for real-time attitude estimation, including factored quaternion algorithm (FQA) for static attitude estimation, and Kalman filtering fordata fusion. A singularity avoidance method, axis-exchanged compensation, is proposed to modify the FQA, allowing the algorithm to track at all attitudes. An R-adapted module is designed to adjust the Kalman gain, which effectively restrains noise due to dynamic nonlinear acceleration, and improves attitude estimation accuracy. Experimental results show that the R-adapted Kalman filter can accurately estimate attitudes in real-time. Additionally, FQA with an axis-exchanged method has good performance in estimating attitudes of singularity points, and the computational efficiency is higher than a previous method by 50%.

Key words: attitude estimation, factored quaternion algorithm (FQA), Kalman filter (KF), micro-electro-mechanical system-inertial measurement unit (MEMS-IMU), singularity compensation