Exploring the Potential of Wrist-Worn Gesture Sensing
MetadataShow full item record
This thesis aims to explore the potential of wrist-worn gesture sensing. There has been a large amount of work on gesture recognition in the past utilizing different kinds of sensors. However, gesture sets tested across different work were all different, making it hard to compare them. Also, there has not been enough work on understanding what types of gestures are suitable for wrist-worn devices. Our work addresses these two problems and makes two main contributions compared to previous work: the specification of larger gesture sets, which were verified through an elicitation study generated by combining previous work; and an evaluation of the potential of gesture sensing with wrist-worn sensors. We developed a gesture recognition system, WristRec, which is a low-cost wrist-worn device utilizing bend sensors for gesture recognition. The design of WristRec aims to measure the tendon movement at the wrist while people perform gestures. We conducted a four-part study to verify the validity of the approach and the extent of gestures which can be detected using a wrist-worn system. During the initial stage, we verified the feasibility of WristRec using the Dynamic Time Warping (DTW) algorithm to perform gesture classification on a group of 5 gestures, the gesture set of the MYO armband. Next, we conducted an elicitation study to understand the trade-offs between hand, wrist, and arm gestures. The study helped us understand the type of gestures which wrist-worn system should be able to recognize. It also served as the base of our gesture set and our evaluation on the gesture sets used in the previous research. To evaluate the overall potential of wrist-worn recognition, we explored the design of hardware to recognize gestures by contrasting an Inertial measurement unit (IMU) only recognizer (the Serendipity system of Wen et al.) with our system. We assessed accuracies on a consensus gesture set and on a 27-gesture referent set, both extracted from the result of our elicitation study. Finally, we discuss the implications of our work both to the comparative evaluation of systems and to the design of enhanced hardware sensing.
Cite this version of the work
Qi Feng Liu (2018). Exploring the Potential of Wrist-Worn Gesture Sensing. UWSpace. http://hdl.handle.net/10012/12913