Sensor Dataglove for Real-time Static and Dynamic Hand Gesture Recognition
Md. Ahasan Atick Faisal, Farhan Fuad Abir, Mosabber Uddin Ahmed
Abstract
Hand gesture recognition has been a widely explored field of Human Activity Recognition (HAR). In this work, we have presented a sensor-based hand gesture recognition framework to classify both static and dynamic hand gestures in real-time using a dataglove that contains a 3-axis accelerometer (ACC), a 3-axis gyroscope, and 5 flex sensors. We have collected data from 35 volunteers performing 14 static and 3 dynamic gestures wearing the dataglove. We have preprocessed the raw flex sensor data using digital filtering techniques and performed mathematical operations on the accelerometer and gyroscope data for determining accurate orientation and motion profile. Four classical machine learning algorithms were used and compared on both datasets. We have achieved maximum accuracy of 99.53% for static gestures and 98.64% for dynamic gestures using the K-Nearest Neighbors (KNN) classifier. Our proposed framework provides real-time wireless hand gesture detection for Human-Computer Interaction (HCI) and Sign Language Recognition (SLR).