Multi-feature based Object Classification using Flexible Gloves inspired by Human Grasping

Abstract

We present high accuracy object classification using flexible gloves and machine learning algorithms. The flexible gloves are designed with two flex sensors mounted on finger joints and two FSR sensors inside fingertips. When grasping an object, electrical signals are acquired from physically deformed sensors. In this paper, the key features of objects are extracted from the mean and standard deviation values of the sensing signal waveforms. We prepared four sets of blocks for classification and each of them had a different size and weight. As a result, we demonstrated the accuracy of the object classification can be achieved 100 % using the multi-featured sensing dataset acquired by the flexible glove. The multi-featured classification method which combines the flexible gloves and machine learning technology shows a great potential application such as visual impairment aid and human-machine interface.