EfficientNetV2-based dynamic gesture recognition using transformed scalogram from triaxial acceleration signal

Abstract In this paper, a dynamic gesture recognition system is proposed using triaxial acceleration signal and image-based deep neural network. With our dexterous glove device, 1D can be measured from each finger decomposed to time-divided frequency components via wavelet transformation, which known as scalogram image-like format. To feed-forward the with single 2D convolutional networks(CNN) allows having temporality easily recognized without any complex such RNN, LSTM, or spatio-temporal feature 3D CNN, etc. classify image general input dimension of RGB channels, we numerically reconstruct fifteen scalograms into one various representation methods. experiments, employ off-the-shelf model, EfficientNetV2 small large model an classification fine-tuning. evaluate system, bulid custom bicycle hand signals dataset under transformation then qualitatively compare reconstruction method matrix addition, use other tools fast Fourier transform, short-time transform explain advantages in terms time-frequency resolution trade-off issue.