Flexible Gesture Recognition Using Wearable Inertial Sensors

This paper proposes a novel glove-based system for hand-gesture recognition. The system tracks fine-grain hand movements using inertial and attitude measurements. Gestures are recognized in real-time by feeding the sensor readings to a machine-learning algorithm. In addition, gestures are communicated wirelessly to external devices for display and control purposes. The machine learning algorithm is based on Linear Discriminant Analysis (LDA), which allows for accurate and low-complexity classification by projecting into a space with improved clustering and reduced dimensionality. The feature vector comprises the angles between each finger relative to the hand palm. A real-time algorithm is developed to ensure features are captured when the gestures are at a steady-state as opposed to gesture transitions. To demonstrate a viable application, the proposed system has been utilized for automatic recognition of American Sign Language (ASL) gestures. As shown in the result section, the system has achieved an accuracy of 85%, and demonstrated flexibility in accommodating new gestures with a new set of training data.