Prospects of using modern machine training methods to improve sitting working conditions through analysis and control of human posture

DOI: 10.31673/2412-9070.2021.042631

Abstract

The article presents an analysis of the latest research conducted to monitor the health of workers with a sedentary lifestyle. The results show that about 75% of all workers in the developed world are in a sedentary position during their professional activities. This leads to long-term disturbances in a person's sitting posture, which in turn provokes musculoskeletal complications in the back, neck, shoulders, arms and legs. Chronic back pain is becoming a daily problem for many people, and in some cases even an occupational disease. It is determined that among the ways to combat the negative effects of sedentary lifestyle and work, one of the main ones is posture control and a cycle of appropriate physical exercises. Both ways require self-control of the individual, for some reminders from different gadgets. In recent years, «smart» devices and ancillary software that accompany modern man in trying to control his health have become more and more developed. One of the areas of research and development in this area is the study of possible ways to use machine learning in this area. The creation of the apparatus of machine learning tools and neural networks has been implemented, the main task of which is the analysis of human posture with images and video streams, output of results in a form acceptable for further use. The goals of the development are to create a system that will control a person during his work, reminding him of his posture, minimizing the potential negative impact on everyone's health. The created system has enough tools for accurate analysis of human posture from a static image and analysis with an accuracy of 92-94% of the video stream. Google Chrome Web Browser extension has been created for the API to use HTML markup language, CSS page style language, and JavaScript based on TensorFlow libraries that import pre-built and trained machine learning systems. This allows the user to control their posture while working on a personal computer at their workplace. The application provides an opportunity to independently monitor a person's posture and report in case of violations.

Keywords: machine learning methods; neural network; convolutional neural network; human posture; sitting method.

References
1. Reinecke S., Hazard R., Coleman K. A continuous passive lumbar motion device to relieve back pain in prolonged sitting // Advances in Industrial Ergonomics and Safety IV. London: Taylor & Francis, 2002. С. 971–976.
2. Winkel K., Jorgensen J. Evaluation of foot swelling and lower-limb temperatures in relation to leg activity during long-term seated office work // Ergonomics. 1986. №29. С. 313–328.
3. Naqvi S. А. Study of forward sloping seats for VDT workstations // Journal of Human Ergology. 1994. №23. С. 41–49.
4. Hoy D., Brooks P., Blyth R. The epidemiology of low back pain // Best Practice & Research: Clinical Rheumatology. 2010. №24. С. 769–781.
5. Kulikajevas A., Maskeliunas R., Damaševičius R. Detection of sitting posture using hierarchical image composition and deep learning. 2021 (PeerJ Computer Science; 7:e442).
6. Kim Y., Son Y., Kim W. Classification of Children’s Sitting Postures Using Machine Learning Algorithms. 2018. 8 с. (MDPI; 1280).
7. Roh J., Park H., Lee K. Sitting Posture Monitoring System Based on a Low-Cost Load Cell Using Machine Learning. 2018. 18 с. (MDPI; 208).
8. Buontempo F. Genetic Algorithms and Machine Learning for Programmers. 2019. 234 с. (The Pragmatic Programmers).