Fingertip-Based Feature Analysis for the Push and Stroke Manipulation of Elastic Objects

In this study, to quantitatively understand finger operations used to manipulate elastic objects, we explore robust fingertip-based feature descriptors that are invariant to operator, finger position, and target object. To measure the tactile information generated when an object is directly touched by a fingertip, we used a wearable system that enables the simultaneous measurement of fingertip position and strain without inhibiting the operator's sense of touch. This paper focuses on the quantitative classification of the push and stroke operations of a single finger, and conducted user experiments to obtain time-series fingertip position and strain from 10 subjects touching nine types of elastic objects. The recognition rate was investigated by binary classification using a support vector machine and cross validation. The results show that the two-dimensional features obtained from fingertip position and strain within a 0.9-s time frame can stably recognize push and stroke operations on elastic bodies of different shapes, stiffnesses, and thicknesses at a higher recognition rate.