A motion recognition algorithm using polytopic modeling
Pierre Moreau, David G. Durand, Jérôme Bosche, Michel Lefranc · 2020
People's movements say a lot about their activities. Whether it concerns sports, music (playing an instrument), at work, in re-education, each domain has its own specific moves. However, some of it, such as sports competition, need high-precision movements. Tools are available to measure movements in all sectors. First, sensors are placed on different strategic points on the person's body that allow us to retrieve temporal data from the body of the user. In this work, no camera is used for motion recognition in order to let the user free to go in different spaces. A considerable number of algorithms help for movement recognition such as deep learning, convolutional neural networks or dynamic modelling, but in most cases, cameras are used. So, our approach consists of two phases. First, model the users thanks to whole-body sensors and save characteristic movements. Second, we still use sensors, to model the test person to find characteristic movements depending on the activity.