Home // SPWID 2018, The Fourth International Conference on Smart Portable, Wearable, Implantable and Disability-oriented Devices and Systems // View article


A Single Wearable IMU-based Human Hand Activity Recognition via Deep Autoencoder and Recurrent Neural Networks

Authors:
Patricio Rivera Lopez
Edwin Valarezo Añazco
Sangmin Lee
Kyung Min Byun
Min Hyoung Cho
Soo Yeol Lee
Tae-Seong Kim

Keywords: Human Hand Activity Recognition; Autoencoder; Deep Learning; RNN; CNN.

Abstract:
Human Hand Activity Recognition (HAR) using wearable sensors can be utilized in various practical applications such as lifelogging, human-computer interaction, and gesture interfaces. Especially with the latest deep learning approaches, the feasibility of HAR in practice gets more promising. In this paper, we present a HAR system based on deep Autoencoder for denoising and deep Recurrent Neural Network (RNN) for classification. The proposed HAR system achieves a mean accuracy of 79.38% for seven complex hand activities, while only of 72.65% without the autoencoder. The presented combination of autoencoder and RNN could be useful for much improved human activity recognition.

Pages: 3 to 7

Copyright: Copyright (c) IARIA, 2018

Publication date: July 22, 2018

Published in: conference

ISSN: 2519-8440

ISBN: 978-1-61208-657-6

Location: Barcelona, Spain

Dates: from July 22, 2018 to July 26, 2018