Hand Gesture Classification using Inaudible Sound with Ensemble Method

Jinwon Cheon, Sunwoong Choi
2020 Advances in Science, Technology and Engineering Systems  
Recognizing the human behavior and gesture has become important due to the increasing use of wearable devices. This study classifies hand gestures by creating sound in the inaudible frequency range from a smartphone and analyzing the reflected signals. We convert the sound using Short-Time Fourier Transform to magnitude and phase. We trained two types of data on Convolutional Neural Network model. And then we propose a method applying soft voting, an ensemble technique, to improve
more » ... accuracy taking the average of two models' result. In this paper, the classification accuracy of the Mag model is 96.0% and the classification accuracy of the Phase model is 90.0% for 8 hand gestures. While the ensemble model showed 96.88%, which is better than Mag and Phase models.
doi:10.25046/aj0506115 fatcat:ddtn672zbnd7zfwt4mpjeduyiu