Dolphin: Ultrasonic-Based Gesture Recognition on Smartphone Platform

User experience of smart mobile devices can be improved in numerous scenarios with the assist of in-air gesture recognition. Most existing methods proposed by industry and academia are based on special sensors. On the contrary, a special sensor-independent in-air gesture recognition method named Dol...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2014 IEEE 17th International Conference on Computational Science and Engineering s. 1461 - 1468
Hlavní autoři: Yang Qifan, Tang Hao, Zhao Xuebing, Li Yin, Zhang Sanfeng
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 01.12.2014
Témata:
Accuracy
Acoustics
Dolphins
Doppler
Doppler shift
Gesture recognition
In-air gesture recognition
Interaction Technique
Manuals
Support vector machine classification
Ultrasonic
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:User experience of smart mobile devices can be improved in numerous scenarios with the assist of in-air gesture recognition. Most existing methods proposed by industry and academia are based on special sensors. On the contrary, a special sensor-independent in-air gesture recognition method named Dolphin is proposed in this paper which can be applied to off-the-shelf smart devices directly. The only sensors Dolphin needs are the loudspeaker and microphone embedded in the device. Dolphin emits a continuous 21 KHz tone by the loudspeaker and receive the gesture-reflecting ultrasonic wave by the microphone. The gesture performed is encoded into the reflected ultrasonic in the form of Doppler shift. By combining manual recognition and machine leaning methods, Dolphin extracts features from Doppler shift and recognizes a rich set of pre-defined gestures with high accuracy in real time. Parameter selection strategy and gesture recognition under several scenarios are discussed and evaluated in detail. Dolphin can be adapted to multiple devices and users by training using machine learning methods.
DOI:10.1109/CSE.2014.273