A Smart Glove Based on Inductive Sensors for Hand Gesture Recognition

This article proposes an enhanced gesture recognition system based on inductive sensors for accurately recognizing American Sign Language (ASL) hand gestures. In this system, conductive threads are employed to sew coils onto a regular glove covering the fingers, wrist, palm, and ulnar positions. To improve sensitivity, a data acquisition system measures a tank circuit formed by each sewn coil and external components. The system is rigorously tested on ten subjects. A comparative study of three machine learning algorithms (MLAs), including random forest (RF), support vector machine (SVM), and K-nearest neighbors (KNNs), is conducted to detect 26 ASL letters. To improve the diversity and generalizability of the MLA, the generative adversarial network (GAN) data augmentation method is provided, expanding the dataset to 5050 trials for each gesture. The results demonstrate impressive accuracy rates of 99.67% and 97.46% using five-fold cross-validation (5F-CV) and leave-one-subject-out cross-validation (LOSO-CV), respectively, for the RF algorithm. This exhibits higher sensitivity in detecting similar gestures compared to the previous design. The proposed solution addresses the limitations of existing hand gesture recognition designs and offers a practical and effective approach to human-computer interaction.