Intelligent Chair Sensor: Classification and Correction of Sitting Posture

Intelligent Chair Sensor: Classification and Correction of Sitting Posture

Leonardo Martins, Rui Lucena, Rui Almeida, João Belo, Cláudia Quaresma, Adelaide Jesus, Pedro Vieira
Copyright: © 2014 |Pages: 16
DOI: 10.4018/ijsda.2014040105
(Individual Articles)
No Current Special Offers


In order to develop an intelligent system capable of posture classification and correction the authors developed a chair prototype equipped with air bladders in the chair's seat pad and backrest, which can in turn detect the user posture based on the pressure inside said bladders and change their conformation by inflation or deflation. Pressure maps for eleven standardized postures were gathered in order to automatically detect the user's posture, with resource to neural networks classifiers. First the authors tried to find the best parameters for the neural network classification of our data, obtaining an overall classification of around 80% for eleven standardized postures. Those neural networks were then exported to a mobile application to achieve a real-time classification of the standardized postures. Results showed a real-time classification of 93.4% for eight standardized postures, even for users that experimented for the first-time our intelligent chair. Using the same mobile application they devised and implemented two correction algorithms, acting due to conformation change of the bladders in the chair's seat when a poor seating posture is detected for certain periods of time.
Article Preview


The last decades have been defined by constant changes in the workplace, transportation, communications, and entertainment in the last century led to a sedentary lifestyle, forcing the population to spend longs periods of time in a sitting position (Chau, et al., 2010; Hartvigsen, Leboeyf-Yde, Lings & Corder, 2000). While seated, most of the bodyweight is transferred to the ischial tuberosities and to the thigh and the gluteal muscles. The rest of the weight is distributed to the ground through the feet and to the backrest and armrest when they are available (Pynt, Higgs & Mackey, 2001). Adopting a lumbar flexion position for long periods of time, leads to a decrease of the lumbar lordosis (van DieËn, De Looze & Hermans, 2001; Cagnie, Danneels, Van Tiggelen, De Loose & Cambier, 2007), which has been linked to back and neck pain due to the anatomical changes of the spine and the degeneration of the intervertebral discs and joints. Adopting a bad posture while seated can worsen these health problems (Lis, Black, Korn & Nordin, 2007). These health problems are one of the leading cause of work-related disability and loss of productivity in industrialized countries (Straaton, Fine, White & Maisiak, 1998) and estimates show that, only in the USA, more than 50$ billion dollars are spent every year for the treatment health problems associated with back pain (DHHS - NIOSH, 2001).

There are a wide number of clinical views of ‘correct’ or ‘incorrect’ postures, but until recent years there were little quantitative studies to define those postures. There have been groups that tried to determine whether the so called ‘good’ postures actually provide a clinical advantage using multiple cameras and analyzed the three-dimensional optical motion of the user (Edmondston, et al., 2007) and using 3-D motion sensors adhered to the skin to measure different spinal angles (Claus, Hides, Moseley & Hodges, 2009).

To solve the problem of incorrect posture adoption for long periods of time in a sitting position, several investigation groups have been using of sheets of surface-mounted pressure sensors arranged in a 2D-array-like manner. These pressure sensors were able to detect the user posture, using the acquired pressure maps and various Classification Algorithms.

Various studies equipped with a sheet array of 42-by-48 pressure sensing elements (one for the seat pad and one for the backrest) were able to distinguish various postures (Slivosky & Tan, 2000; Tan, Slivovsky & Pentland, 2001; Zhu, Martinez, & Tan, 2003). Both Slivosky and Tan (2000) and Tan, Slivovsky, and Pentland (2001) used Principal Component Analyses (PCA) for posture detection for human-machine interactions obtaining an overall classification accuracy of 96% and 79% for familiar and unfamiliar users, respectively, while Zhu, Martinez and Tan (2003) used the same data acquisition methods from the previous two studies to investigate which classification algorithms would be the best for static posture classification. These authors found that among k-Nearest Neighbor, PCA, Linear Discriminant Analysis and Sliced Inverse Regression (SIR), both PCA and SIR outperformed the other methods (Zhu, Martinez & Tan, 2003). One group, also using the same sensor sheets, studied the relationship between patterns of postural behaviors and interest levels in children (Mota & Picard, 2003). They were able to classify nine postures in real time achieving an overall accuracy of 87.6% with new subjects, while using Hidden Markov Models to study the interest levels while performing a learning task on a computer (Mota & Picard, 2003). Another group used two sheet array of 32-by-32 sensors (for the seat pad and backrest) for a car driver biometric identification system (Riener & Ferscha, 2008).

Complete Article List

Search this Journal:
Volume 12: 1 Issue (2024): Forthcoming, Available for Pre-Order
Volume 11: 5 Issues (2022)
Volume 10: 4 Issues (2021)
Volume 9: 4 Issues (2020)
Volume 8: 4 Issues (2019)
Volume 7: 4 Issues (2018)
Volume 6: 4 Issues (2017)
Volume 5: 4 Issues (2016)
Volume 4: 4 Issues (2015)
Volume 3: 4 Issues (2014)
Volume 2: 4 Issues (2013)
Volume 1: 4 Issues (2012)
View Complete Journal Contents Listing