Signal processing and machine learning for real-time classification of ergonomic posture with unobtrusive on-body sensors; application in dental practice

Over 80% of dentists report having some type of back, neck or shoulder pain. Research has identified significant costs linked to a very high rate of Work-Related Musculoskeletal Disorders (WMSDs) associated with poor ergonomic positioning in dentists. The annual costs of WMSDs across all occupations are estimated to be between 13 and 54 billion dollars. Little research has been done to explore the design of portable, inexpensive, non-invasive and unobtrusive real-time systems to measure posture. This paper details the design and testing of our proposed system that applies signal processing and robust machine learning techniques to improve the ergonomics of dental practitioners. We outline a number of different signal processing and classification techniques tested and analytically compared with our proposed system. Our system makes use of commercial inclinometers embedded into a standard laboratory coat. The ability of the system to measure posture accurately in practical settings, without needing exact and obtrusive placement of sensors or extensive calibration, is demonstrated through a set of experiments with human subjects.