The micro-electro-mechanical technology facilitates the development of low-cost orientation estimation solutions using inertial and magnetic sensors. However, the magnetic disturbances challenge the robust pitch and roll measurements on some occasions, such as the quadrotor, ship, and railway. This work analyzes the coupling mechanism of the magnetic information on the roll and pitch estimates. Subsequently, it proposes a quaternion-based decoupled sensor fusion approach under the frame of an extended Kalman filter by taking the orthogonal gravitational vector and horizontal magnetic vector as absolute observation references. Three experimental tests validated its performances in robust roll and pitch measurements under magnetic disturbances, static and dynamic robotic motion estimation, and human upper body pose tracking. Results show that the proposed sensor fusion approach could keep the roll and pitch immune to the magnetic disturbances, and provide reliable static and dynamic orientation estimation accuracy. Moreover, it is promising to provide feedback for rehabilitation training.

微机电技术有助于开发使用惯性和磁传感器的低成本方向估计解决方案。然而，磁干扰在某些情况下对稳健的俯仰和横滚测量提出了挑战，例如四旋翼、轮船和铁路。这项工作分析了磁信息对横滚和俯仰估计的耦合机制。随后，在扩展卡尔曼滤波器框架下，以正交引力矢量和水平磁矢量为绝对观测参考，提出了一种基于四元数的解耦传感器融合方法。三项实验测试验证了其在磁干扰下稳健的横滚和俯仰测量、静态和动态机器人运动估计以及人体上身姿势跟踪方面的性能。结果表明，所提出的传感器融合方法可以保持滚动和俯仰不受磁干扰的影响，并提供可靠的静态和动态方向估计精度。此外，它有望为康复训练提供反馈。