Previous Abstract Return to Session A4 Next Abstract

Session A4: Integrated Inertial Navigation Systems

Optically Enabled Multi-Sensor Automatic Calibration of Magnetic Sensors
Stephen Pledgie, Oleg Nesterenko, Pete Tewksbury, Nick Orange, Jamie Marraccini, Inertial Labs
Location: Pavilion Ballroom East
Date/Time: Wednesday, Apr. 22, 4:52 p.m.

Magnetic field sensors (magnetometers) are frequently used as a cost-effective, reliable means to determine an object’s three-dimensional (3D) orientation relative to the Earth’s magnetic north pole. An object’s 3D position can also be determined through magnetic field sensing and localization within a magnetic field map. Magnetometers must, however, be properly characterized and calibrated so that the Earth’s magnetic field direction can be accurately inferred despite the presence of disturbances in the local magnetic field created by nearby ferromagnetic materials, including, for instance, those present in electronics and various mechanical mounts. Calibration is also used to determine the specific alignment, in this case relative rotation(s), between multiple single or multi-axis magnetometers themselves and between magnetometers and other sensors, such as accelerometers and gyros. Calibration procedures typically involve moving and rotating a sensor through a predetermined sequence of orientations, oftentimes with the help of a rig, while acquiring and subsequently processing data from magnetic, inertial, and rotational sensors. To the extent that ferromagnetic properties of the device or environment change over time, the calibration procedure must be repeated oftentimes at great cost or impact to the overall system. There is, therefore, a general need for robust, automatic calibration of magnetic sensors that can be performed online, as a background process, using the most reliable reference data available at a particular moment.

In this paper, we describe a capability for making real-time (online) adjustments to magnetic calibration parameters using optically-derived sensor orientation information to enhance and oftentimes replace that available from inertial and other configuration sensors. Once initialized, the optical sensor pipeline transforms camera data (viewed objects, features, and landmarks) into estimates of absolute (global, navigation frame) camera orientation and relative changes (since start or since global reset) in camera orientation and that are subsequently quality-audited and used with 3D magnetic field measurements to compute corrections to soft-iron and hard-iron magnetometer calibration parameters. If optically-derived orientation information becomes unavailable, inertially-derived orientation information is used in its place over short periods of time. Multi-sensor cross-validation and single-sensor measures of usability (MOU) enable automated selection of best orientation reference data throughout the device calibration process and even during device field operation. The capability to operate continuously as a background monitoring and corrective process enables a high degree of adaptation in nonstationary environments featuring magnetic interference and in situations where the magnetic properties of the device under measure may change due to its application utilization. The ability to reliably detect and understand if and when the device under measure changes its magnetic properties makes the optically enabled multi-sensor calibration method described herein unique.

In what follows, the process of magnetic sensor calibration within a navigation filtering framework is first introduced. We then make the case for use of optical sensing as a means for achieving robust, “free form,” continuous calibration of magnetometers in both controlled (laboratory) and open (field) environments. Next, an optically enabled multi-sensor calibration framework is presented and mathematically modeled. Several methods of calibration data processing are detailed, including linear quadratic filtering, recursive least squares, and chunk-based batch optimization. Experimental methods are described and calibration results under different conditions and processing modes are presented and critically discussed with an eye towards practical implementation in product-oriented applications. We conclude with scientific and technical recommendations for future work.

Previous Abstract Return to Session A4 Next Abstract