Advancing Human Lower-limb Kinematic Estimation Using Inertial Measurement Units

Abstract

The study of human biomechanics has broad applications in human health, worker safety, warfighter performance, athlete performance, injury prevention, and related fields. Historically, research in all of these fields is frequently limited by measurements of human kinematics being restricted to laboratory environments. Wearable sensors, in the form of body-worn inertial measurement units (IMUs), show great promise in extending the validity of research conclusions by enabling measurements in non-laboratory environments such as the workplace, home, clinic, and training facility. However, to accurately estimate human kinematics from body-worn IMUs, advancements must be made in signal processing methods to correct integration drift errors caused by the integration of noisy sensor data. This dissertation addresses this need by contributing a novel error-state Kalman filter (ErKF) method for estimating the kinematics of the human lower limbs in broad contexts. The lower limbs are chosen due to their paramount importance in the applications articulated above. This research achievement follows the systematic progression of three studies that advance IMU-based kinematic estimation for: 1) a single foot-mounted IMU, 2) an array of three body-worn IMUs in a mechanical “walker” (an approximation to the human lower limbs), and 3) an array of seven body-worn IMUs in a full representation of the human lower limbs. The major findings and contributions of each study are summarized below.

The first study lays a critical foundation for the full lower-limb model by exploring the limiting case of deploying a single foot-mounted IMU to estimate foot trajectories. This study contributes criteria for selecting IMU sensor hardware to achieve accurate estimates of stride parameters (e.g., stride length, stride angle) and reveals that prior zero-velocity drift corrections developed for normal walking remain applicable for highly dynamic gaits, including fast walking and running.

The second study builds from the first by considering three IMUs attached to the three segments of a mechanical “walker” (composed of a pelvis and two straight legs) which serves as an approximation to the human lower limbs. The study contributes a novel ErKF method to estimate the kinematics of the coupled, three-body walker model. Importantly, the method uses kinematic constraints to reduce integration drift errors without reliance on magnetometers or common assumptions (e.g., level-ground). The method successfully estimates the kinematics of a mechanical walker which replicate closely those obtained via simulation and experimental motion capture (MOCAP). For instance, the (hip) joint angles achieve RMS differences below 1.5 degrees compared to MOCAP.

The success of the ErKF method on the three-body walker model motivates its extension to a full, seven-body model of the human lower limbs in the third study. This study contributes novel joint axis corrections within the ErKF for the hip and knee to reduce joint angle drift errors and to account for the additional complexities of human anatomy (e.g., soft tissue, biological joints). The resulting full model is evaluated on human subjects performing six different types of gait and compared to results from MOCAP. This comparison reveals RMS differences in joint angle estimates generally below 5 degrees when compared to MOCAP employing reflective markers attached to the IMUs. Similarly, small differences in the estimated joint angle ranges of motion, stride length, and step width confirm the significant promise of this novel ErKF method as a research strategy for non-laboratory based biomechanical studies of the human lower limbs and in broad contexts.