Wearable Smart Rings for Multi-Finger Gesture Recognition Using Supervised Learning

This paper presents a wearable smart ring with an integrated Bluetooth low-energy (BLE) module. The system uses accelerometer and gyroscope to collect fingers’ motion data. A prototype was manufactured, its performance tested. To detect complex finger movements, two rings are worn on the point thumb fingers while performing gestures. Nine pre-defined movements were introduced verify feasibility of proposed method. Data pre-processing techniques, including normalization, statistical feature extraction, random forest recursive elimination (RF-RFE), k-nearest neighbors sequential forward floating selection (KNN-SFFS), applied select well-distinguished vectors enhance gesture recognition accuracy. Three supervised machine learning algorithms used for classification purposes, namely Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Naive Bayes (NB). We demonstrated that when utilizing KNN-SFFS recommended features as input, our approach not only significantly decreases dimension vector, results in faster response time, prevents overfitted model but also provides approximately similar prediction accuracy compared all elements used. Using KNN primary classifier, can accurately recognize six one-finger three two-finger gestures 97.1% 97.0% accuracy, respectively.