Attitude estimation of a wearable motion sensor

This paper is concerned with the problem of attitude estimation for a wrist-worn motion sensor equipped with a 3-axis gyroscope and a 3-axis accelerometer. In the absence of motion and the with the arm in its natural resting state, the attitude is unobservable because the accelerometer's measurement of the gravity vector alone can not distinguish between configurations of the motion sensor obtained by rotating the sensor around the wrist. In the presence of motion, considered here to be the swinging motion of the arm, the observability is shown to improve. The estimation task is mathematically formulated as a continuous-time filtering problem on the product Lie group SO(3)×SO(2). The feedback particle filter (FPF) algorithm is explicitly constructed in this setting, using both the rotation matrix and the quaternion. Experimental results with real sensor data are provided to illustrate the tracking performance of the proposed filter.