Smartphone-compatible robust classification algorithm for the Tongue Drive System

In this paper we present several improvements in the Tongue Drive System magnetic sensor signal processing algorithm and its new 8-command classification performance by applying robust pattern recognition techniques that are resistant to natural tongue movements and earth´s magnetic field variations. Two additional commands have been added to the system for better compatibility with smartphone interfaces. For this algorithm support vector machine with radial basis function kernels has been proposed instead of the previous algorithm that was based on linear distance discrimination. The proposed algorithm has also been compared with two commonly used nonlinear classifiers, the Neural Network and AdaBoost. Data was collected from six able-bodied subjects while they sat, walked and read a passage. The average false positive ratio while walking and talking was 2.73% error, and the overall system accuracy under normal conditions was 92.64% correctness, 2.87% misclassification error, and 4.49% rejection error.