A Multivariate Time-Warping Based Classifier for Gesture Recognition with Wearable Strain Sensors

Conductive elastomer elements can be industrially embedded into garments to form unobtrusive strain sensing stripes. The present article outlines the structure of a strain-sensor based gesture detection algorithm. Current sensing prototypes include several dozens of sensors; their redundancy with respect to the limb's degrees of freedom, and other artifacts implied by this measurement technique, call for the development of novel robust multivariate pattern-matching techniques. The algorithm's construction is explained, and its performances are evaluated in the context of motor rehabilitation exercises for both two-class and multi-class tasks.