More about HKUST
Design and Analysis of Natural Hand-Gesture Interfaces for Extended Realities
PhD Thesis Proposal Defence Title: "Design and Analysis of Natural Hand-Gesture Interfaces for Extended Realities" by Mr. Kirill SHATILOV Abstract: Augmented, Mixed, and Virtual Realities, which are collectively referred to as Extended Realities (XR), present significant challenges to conventional input and output methods. Traditional interfaces such as touch screens, keyboards, pointers, and controllers often fall short of meeting the requirements for interacting with both real and virtual objects. They also struggle to provide the necessary mobility, efficiency, and social acceptance that users demand. In this thesis, we aim to design and evaluate innovative natural hand-gesture interfaces tailored for extended realities. To begin with, we explore the use of surface electromyography (sEMG), a non-intrusive wearable imaging technique, for recognizing gestures. sEMG works by detecting electrical signals generated by muscles through the skin's surface, which can then be interpreted to identify performed or intended gestures. In our initial study, we developed and assessed a gesture recognition system intended for controlling a 3D-printed prosthetic hand. Our primary focus is on the mobility of the solution, where deep learning algorithms for gesture recognition are executed on a mobile device, with the option to offload computations to a remote server. In this study, we design the prosthesis, which includes the necessary electronics and a modified chassis, as well as a mobile companion device and communication protocols that link the system's components. Our findings demonstrate that a low-cost, mobile-centered system can achieve state-of-the-art accuracy in gesture recognition while maintaining reliable operation over extended periods. We evaluate the accuracy of gesture recognition across various gesture sets and assess the system's response time and power consumption. Building on the insights gained from our first study, we adapt our gesture recognition system to function as a virtual keyboard for extended realities. In this system, users select keys on a standard QWERTY keyboard by orienting their forearm in space to choose a column (measured by an IMU sensor) and then selecting a specific key within that column through a directional gesture (utilizing the sEMG sensor). Notably, we consider three different usage scenarios for our proposed solution: one with an empty hand, and two with a busy hand — such as when holding an umbrella (cylindrical grasp) or a pen (tripod grasp). This approach results in an input system for extended realities that offers an uninterrupted experience for subtle and even discreet text entry. We conduct experiments involving a dozen participants to evaluate text input rates, identify common errors, and assess the overall usability of our proposed text input system. Date: Thursday, 29 May 2025 Time: 2:00pm - 4:00pm Venue: Room 2128C Lift 19 Committee Members: Prof. Pan Hui (Supervisor, EMIA) Dr. Tristan Braud (Co-supervisor) Prof. Pedro Sander (Chairperson) Prof. Gary Chan