Affiliations: [a] School of Information Science and Engineering, Lanzhou University, Lanzhou, China. E-mails: [email protected], [email protected], [email protected], [email protected] | [b] Ministry of Education, Key Laboratory of Dependable Service Computing in Cyber Physical Society, Chongqing, China | [c] School of Big Data & Software Engineering, Chongqing University, Chongqing, China. E-mail: [email protected]
Abstract: Wearable device is becoming more and more popular, the emergence of wearable equipment, is widely used in daily life, medical, industrial and scientific research and so on. But most of the wearable equipment still needs people’s actual operation, which will bring a lot of inconvenience. For example, when people’s hands are not available, they can’t be operated. Now, many sensors are embedded in off-the-shelf smartwatch, which make it possible to detect different finger gestures by using these sensors. Past research mostly concentrates on identifying gross hand gestures, which usually need the help of other additional hardware that is expensive and discomfort. In this work we utilize muscle activity data getting from integrated motion sensors (accelerometer, linear accelerometer and gyroscope) nested in smartwatch to recognize tiny finger gestures. We extract features within a 1 second sliding window. We test classification effect of different classifiers and different sensors. In the result, we can recognize 5 unremarkable finger gestures well and identify specific finger gestures with a high accuracy. The result can be applied on current popular smartwatches interaction. Also our result will lay the foundation for related research in finger gestures recognition field.
