PROJECT TITLE :

Push the Limit of Acoustic Gesture Recognition

ABSTRACT:

The proliferation of smart devices and the applications built for them has led to an increased interest in controlling devices through the use of gestures. This interest has expanded to include ubiquitous sensing and interaction. Recent research has focused on recognizing hand gestures and tracking hand movement through the use of acoustic signals. However, because of frequency selective fading, interference, and a lack of adequate training data, they have a low robustness. We propose RobuCIR, a robust contactless gesture recognition system, as a result of this body of work. RobuCIR is able to function accurately and robustly under a variety of different practical impact factors. RobuCIR uses a mechanism called frequency-hopping in order to prevent signal interference and reduce the effects of frequency selective fading. We are investigating a number of data augmentation techniques based on a limited volume of collected data in order to simulate a variety of different practical impact factors in order to further improve the robustness of the system. The augmented data is utilized to effectively train neural network models and to contend with a wide variety of influential factors (e.g., gesture speed, distance to transceiver, etc .). The findings of our experiments indicate that RobuCIR is capable of recognizing 15 different hand gestures and outperforms other works that are considered to be state-of-the-art in terms of accuracy and robustness.