Towards grip sensing for commodity smartphones through acoustic signature

While hand grips are important to understand the intent of smartphone users, existing studies on hand grip detection either require additional hardware or exhibit limitations on the type/number of grips. In this paper, we propose a novel grip sensing system that enables a smartphone to detect various user-defined hand grips without any additional hardware. Our system emits a carefully-designed (inaudible) sound signal, and records the sound signal modified by an individual grip. The recorded sound signal is transformed into a unique sound signature through feature extraction process, and then SVM (Support Vector Machine) classifies the sound signature so as to identify the signature as one of pre-defined grips. With six representative grips, we demonstrate that our system exhibits 93.0% average accuracy for ten different users. Beyond this feasibility demonstration, our ongoing work is not only to improve the accuracy, but also to adapt our system to various real environments.