Wearable Multisensory Glove for Shape, Size, and Stiffness Recognition Based on Off-the-Shelf Components

Abstract

This letter presents a wearable multisensory glove that integrates commercial sensors, off-the-shelf components, and an embedded machine learning (ML) approach for object recognition. Sixteen printed objects, categorized by shape, size, and stiffness, were examined using the developed system. Time-domain features and raw data fed ML algorithms including single-layer feed-forward neural network, multilayer perceptron (MLP), and 1-D convolution neural network (1D-CNN). The algorithms were deployed on a low-cost Arduino Nano 33 BLE sense edge device for real-time recognition. Results demonstrate that 1D-CNN achieved the highest classification accuracy of 99.2%, with an inference time of 167 ms while consuming only 2.8 mJ of energy per inference. This study demonstrates the effectiveness of the proposed system in recognizing objects opening up interesting perspectives for various biomedical applications, such as poststroke rehabilitation.