The Feasibility of Recognizing Pinch Gestures with Commodity Smartwatch Hardware and Machine Learning
Christopher Kinzel and Anthony Tang. (2017). The Feasibility of Recognizing Pinch Gestures with Commodity Smartwatch Hardware and Machine Learning. In Poster Proceedings of Graphics Interface 2017. Notes: 2-page abstract + poster.
Lacking the large screen size of mobile phones, smartwatch interaction faces challenges with regards to finger occlusion and dense hard to target controls that arise from their extremely tiny screens and input surfaces. Attempts to reconcile this problem often burden the user with bulky, expensive, extra hardware and lack simplicity and accuracy. Our approach uses a set of pinching gestures between the thumb and individual fingers on the smartwatch hand to provide a simple, socially acceptable, and comfortable input method. Our implementation makes use of common onboard sensors and machine learning to process the noisy, complex sensor data. Our method can be used to augment and extend traditional touch screen input on smartwatch devices. We evaluate our approach by measuring classification accuracy in a small user study with two different scenarios one where the user is stationary and one involving movement. Our results indicate that the approach is a feasible extension to touchscreen interaction but further work is needed to increase classification during movement.