Neural Network Model for Electromagnetic Position Tracker Recalibration

Abstract

Besides the several benefits of electromagnetic trackers, these devices have an important limitation, they are greatly affected by the magnetic disturbances in the environment. For this reason, in the past, several studies have developed different calibration techniques to improve the accuracy of electromagnetic trackers. Past studies concluded that a Neural Network based calibration produced better results than other commonly used calibration techniques. In those studies, the data was either collected manually using a regularly spaced grid with known coordinates or using another motion capture technique such as optical tracking. However, the first acquisition method is time consuming and provides a small dataset for the training of a Neural Network. On the other hand, the second method requires the use of an optical tracking system for every calibration. Here, we present a recalibration technique that uses a combination of both acquisition methods to reduce the error of the tracked position. First, a Neural Network is trained using a big dataset obtained with an optical motion tracking system.