Enhanced gesture sensing using battery-less wearable motion trackers

• Main Author: TRAN, Huy Vu
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2019
• Subjects: Gesture; Hand Tracking; Low-latency; Battery-less; WiFi; Energy Harvesting; Beam-forming; Software Engineering
• Online Access: https://ink.library.smu.edu.sg/etd_coll/251; https://ink.library.smu.edu.sg/cgi/viewcontent.cgi?article=1251&context=etd_coll
• Institution: Singapore Management University

Wearable devices are gaining in popularity, but are presently used primarily for productivity-related functions (such as calling people or discreetly receiving notifications) or for physiological sensing. However, wearable devices are still not widely used for a wider set of sensing-based applications, even though their potential is enormous. Wearable devices can enable a variety of novel applications. For example, wrist-worn and/or finger-worn devices could be viable controllers for real-time AR/VR games and applications, and can be used for real-time gestural tracking to support rehabilitative patient therapy or training of sports personnel. There are, however, a key set of impediments towards realizing this vision. State-of-the-art gesture recognition algorithms typically recognize gestures, using an explicit initial segmentation step, only after the completion of the gesture, thereby being less appropriate for interactive applications requiring real-time tracking. Moreover, such gesture recognition & hand tracking is relatively energy-hungry and requires wearable devices with sufficient battery capacity. Such battery-driven operation further restricts widespread adoption, as (a) the device must be periodically recharged, thereby requiring human intervention, and (b) the battery also adds to the wearable device’s weight, which potentially affects the wearer’s motion dynamics. In this thesis, I explore the development of new capabilities in wearable sensing along two different dimensions which we believe can help increase the diversity and sophistication of applications and use cases supported by wearable- based systems: (i) Low-latency, low-complexity gesture tracking, and (ii) Ultra-low-power or Battery-less operation. The thesis first proposes the development of a battery-less wearable device that permits tracking of gestural actions by harvesting power from appropriately beamformed WiFi signals. This work requires innovations in both wearable and WiFi AP operations, which work together to support adequate energy harvesting over distances of several meters. Through a combination of simulations and real-world studies, I show that (a) smart WiFi beamforming techniques can help support sufficient energy harvesting by up to 3-4 battery-less devices in a small room, and (b) the prototype battery-less wearable device can support uninterrupted tracking of significant gestural activities by an individual. The thesis then explores the ability of smartwatch to recognize hand gestures early and to track the hand trajectory with low latency, so that it can be used in realizing interactive applications. In particular, I show that our techniques allow a wrist-worn device to be used as a real-time hand tracker and gesture recognizer for an interactive application, such as Table Tennis. The dissertation also demonstrates that my proposed method provides a superior energy-vs-accuracy trade-off compared to more complex gesture tracking algorithms, thereby making it more conducive to operation on battery-less wearable devices. Finally, I evaluate whether my proposed techniques for low- latency gesture recognition can be supported by WiWear-based wearable devices, and establish the set of operating conditions under which such operation is feasible. Collectively, my work advances the state-of-the-art in low-energy wearable-based low-latency gesture recognition, thereby opening up the possible use of battery-less, WiFi-harvesting based devices for gesture-driven applications, especially for sports & rehabilitative training.