Effects of Outlier Elimination in Magnetometer Calibration Data on the Performance of Ellipsoid Fitting-based Calibration

A magnetometer should be calibrated to correct errors such as offset, scale factor, misalignment soft iron distortion, and hard iron distortion so as to achieve accurate azimuth estimation. An outlier elimination method has been recently proposed to improve calibration performance by eliminating the outliers of calibration data. In this method, it is assumed that calibration data is collected in a uniform distribution to form a sphere. However, considering the actual data collection process, it is difficult to guarantee uniformly distributed data. Thus, this study investigates the effects of outlier elimination on the performance of ellipsoid fitting-based magnetometer calibration. In this study, the calibration performances before and after outlier elimination were compared for the following three cases: uniformly distributed data with insignificant outliers (Case 1), uniformly distributed data with significant outliers (Case 2), and nonuniformly distributed data (Case 3). In addition, the calibration performances were compared according to a threshold for detecting outliers. The results of this study showed that outlier elimination changed the azimuth estimation performance by -0.002°, 0.711°, and -2.588° for Cases 1–3, respectively, in terms of root mean squared error. Overall, these results indicate that outlier elimination is effective for uniformly distributed data with significant outliers; however, it causes performance degradation for nonuniformly distributed data. Therefore, it is essential to examine the distribution of calibration data before applying outlier elimination.