Magnetometer-Free IMU-Based Joint Axis Calibration and Estimation

According to anatomic equivalent relation, the position of the joint center cannot be measured directly, so calibration of the joint attaches great significance to human motion analysis. Several algorithms have coped with it, however, algorithm complexity and calibration speed with accurate human model make it tough to widely apply. Hence, a modified calibration method is proposed to deal with these problems, where none of jigs, other equipment, or specified action are required in the calibration process. As the main contribution, the robustness and convergence rate are increased while the error in calibration with data of IMU is reduced, and Levenberg-Marquart method is used to calculate joint axes and minimize error. Compared to Gauss-Newton, Levenberg-Marquart has strong convergence and robustness, even if the initial position value is far from the actual position or if the determinant matrix is close to zero. Subsequently, a polynomial interpolation compensates the error caused by the serrated points. Finally, an experiment makes validation of this method. The result indicates that the algorithm converges within four iterations and the error is almost close to zero. Moreover, IMU can be installed arbitrarily since magnetometer-free. Online pre-processing of data and smoothing of anomalous velocity sawtooth points allows the IMU to be easily applied to exoskeletons and human motion intent recognition.