Improving the Accuracy of Human Body Orientation Estimation With Wearable IMU Sensors

Accurately estimating the orientation of different human body segments using low cost inertial sensors is a key component in various activity-related and healthcare-related applications. Typically, the signals from a gyroscope and an accelerometer are fused inside a Kalman filter to determine the orientation. However, the accelerometer measurements are influenced by the linear accelerations of the body segments in addition to the gravitational acceleration that corrupts the orientation estimates. The conventional method to deal with linear acceleration is to model it as a first-order low-pass process and estimate it inside the Kalman filter. In this conventional method, important information from those sensor axes that do not experience linear accelerations is lost. In this paper, we modify the conventional approach to deal with the problem of linear acceleration more efficiently. The proposed approach estimates the direction of linear acceleration and assigns lower weights inside the Kalman filter to only those sensor axes that are experiencing acceleration, thus conserving important information from other axes measurements. The proposed method is compared with the conventional method using simulations and experimentation on a test subject performing daily routine tasks. The results indicate a significant performance improvement in orientation estimation.