Closed-chain pose estimation from wearable sensors

Inertial measurement unit sensors are commonly used for human pose estimation. However, a systematic and robust method to incorporate position and orientation constraints in the kinematic structure during environmental contact is lacking. In this paper, we estimate the pose using the extended Kalman filter, linearize the closed loop constraints about the predicted Kalman filter state, then project the unconstrained state estimate onto the constrained space. Multiple constraints that are representative of real world scenarios are derived. The proposed technique is tested on two human movement datasets and demonstrated to outperform unconstrained Kalman filter.