Improving the accuracy of wearable sensor orientation using a two-step complementary filter with state machine-based adaptive strategy

Magnetic and inertial sensors have been widely used in human motion analysis, where accurate orientation estimation is important. Regardless of methods, sensor fusion process faces two major challenges: external acceleration and magnetic disturbance. By analyzing the existing literature, we found that a suitable base sensor fusion algorithm integrated with proper adaptive strategies is essential. In this paper, we first implemented a quaternion-based two-step complementary filter as the base sensor fusion algorithm. Its attitude estimation is immune to magnetic disturbance, and it contains two separate tuning parameters for different conditions of external acceleration and magnetic disturbance. With this base algorithm, we proposed a novel finite state machine-based adaptive strategy. Two state machines were developed to cope with external acceleration and magnetic disturbance. To validate the performance of the proposed sensor fusion method systematically, we developed a battery of tests representing daily-living environments, including acceleration distorted condition, magnetically distorted condition, and combined distorted condition. Also, a real-world experiment was performed to validate the orientation estimation accuracy on foot trajectory calculation. The results demonstrate that the proposed sensor fusion method performs well against external acceleration and magnetic disturbance compared with the existing methods. Especially, the proposed sensor fusion method showed a very high accuracy in the 60 s continuous combined distorted condition, where the root mean square errors of the roll, pitch and yaw were 0.63°, 0.83° and 0.96°, respectively. The accuracy of foot trajectory estimation was significantly improved with the orientation estimated by the proposed method. In conclusion, the proposed sensor fusion method is encouraging for human motion-related applications in daily-living environments. In addition, the proposed state machine based adaptive strategy is simple and robust, and can be easily integrated into other base sensor fusion algorithms.