Motivation behind the research
Bad postures adopted in the workplace can have a harmful impact on workers leading to injuries in the long term. We thought about implementing a solution combining sensors and Artificial Neural Networks to detect awkward postures before any injury appears so that we can immediately alert workers in case the posture is harmful.
Among the parameters we used to analyze the posture, we also included Center of Pressure (COP). This represents the sum of all forces acting between the human body and the ground on which they are standing. It is a largely used parameter in posture studies.
We classified postures into one of six predefined classes of postures. We first defined three common handling tasks by workers, each of them has an acceptable version and an awkward one. We trained the algorithm by repeating several times the six postures.
The training process allows the system to learn how to differentiate between each of the postures, and involves three main steps: The acquisition phase, the preprocessing phase and the classification phase.
- The acquisitions phase
In this phase, we recorded the data coming from the sensors. The sensors we used were: an accelerometer placed in the helmet  to calculate the head accelerations, and four force sensors installed in the insole (shoe) . It’s with the force sensors that we were able to calculate the trajectory of the Center of Pressure (COP).
- The preprocessing phase
During this phase, we extracted certain important features from our recorded signals. For e.g. using the data from force sensors, we were able to calculate the trajectory of the COP, and from the trajectory, we were able to extract the area of the COP displacement. After we extracted all the features, we reduced them as much as possible such as to keep only the best ones that can differentiate between the different classes.
- The classification phase
This phase involved using a classifier- an Artificial Neural Network in our case. To make it simple, you can imagine it as a black box that takes the current posture as an input, and tells you whether it is good or bad.
One of the most interesting outcome of our study is the type of features we used. Along with the classic features used by many others, we used new Center of Pressure (COP) features which improved our classification accuracy to 95%.
The main limitation is that we only used one participant. But since the system is supposed to be used by one worker and configured using his measurements, this can be considered as a fair initial prototype.
Our system is aiming at preventing the appearance of MSDs at the workplace. We are still on our first prototype, and we are working on scaling and improvements.
Research Article: Wearable Devices for Classification of Inadequate Posture at Work Using Neural Networks, Sensors 2017.
 Li, P.; Meziane, R.; Otis, M.J.-D.; Ezzaidi, H.; Cardou, P. A Smart Safety Helmet using IMU and EEG sensors for worker fatigue detection. In Proceedings of the IEEE International Symposium on RObotic and SEnsors Environments (ROSE), Timisoara, Romania, 16–19 October 2014.
 Otis, M. J.-D., Ayena, J. C., Tremblay, L. E., Fortin, P. E., & Ménélas, B.-A. J. (2016). Use of an Enactive Insole for Reducing the Risk of Falling on Different Types of Soil Using Vibrotactile Cueing for the Elderly. PloS One, 11(9)
Source: The Surg