Detecting tapping motion on the side of mobile devices by probabilistically combining hand postures

We contribute a novel method for detecting finger taps on the different sides of a smartphone, using the built-in motion sensors of the device. In particular, we discuss new features and algorithms that infer side taps by probabilistically combining estimates of tap location and the hand pose--the hand holding the device. Based on a dataset collected from 9 participants, our method achieved 97.3% precision and 98.4% recall on tap event detection against ambient motion. For detecting single-tap locations, our method outperformed an approach that uses inferred hand postures deterministically by 3% and an approach that does not use hand posture inference by 17%. For inferring the location of two consecutive side taps from the same direction, our method outperformed the two baseline approaches by 6% and 17% respectively. We discuss our insights into designing the detection algorithm and the implication on side tap-based interaction behaviors.