Home // HEALTHINFO 2019, The Fourth International Conference on Informatics and Assistive Technologies for Health-Care, Medical Support and Wellbeing // View article
Authors:
Christopher Pramerdorfer
Martin Kampel
Johannes Heering
Keywords: Workplace health promotion, Sitting posture estimation, Deep learning, Depth data analysis
Abstract:
Prolonged sitting in an unhealthy posture is a common cause of back pain and other health problems in office workers. People are often not aware that they are sitting in an unhealthy way, the problems this can cause in the long term, and how they should improve their posture. We present a system that is able to provide this information by analyzing people's postures over several days. The system is fully automatic and requires no worn devices. Instead, data from a depth sensor is used for periodic 3D upper-body pose estimation. This pose estimation is carried out by a convolutional neural network that was trained on synthetic depth data to overcome the lack of available real-world datasets. On this basis, each pose is assigned to one of several common classes of healthy and unhealthy sitting poses. This results in a large collection of body poses and classification results, which are used to generate a personalized posture report that includes suggestions for improving the sitting posture. We show experimentally that the system is able to estimate 3D poses and perform pose classification with high accuracy.
Pages: 49 to 53
Copyright: Copyright (c) IARIA, 2019
Publication date: November 24, 2019
Published in: conference
ISSN: 2519-8491
ISBN: 978-1-61208-759-7
Location: Valencia, Spain
Dates: from November 24, 2019 to November 28, 2019