Quantitative assessment of magnetic sensor signal processing algorithms in a wireless tongue-operated assistive technology.

In this paper, we evaluate the overall performance of various magnetic-sensor signal processing (mSSP) algorithms for the Tongue Drive System based on a comprehensive dataset collected from trials with a total of eight able-bodied subjects. More specifically, we measure the performance of nine classifiers on the two-stage classification used by the mSSP algorithm, in order to learn how to improve the current algorithm. Results show that is it possible to reduce misclassification error from 5.95% and 20.13% to 3.98% and 5.63%, from the two assessed datasets, respectively, without sacrificing correctness. Furthermore, since the mSSP algorithm must run in real time, the results show where to focus the computational resources when they are constrained by the platforms with limited resources, such as smartphones.