Dance Gestures Recognition for Wheelchair Control

Abstract

Wheelchair dance is an inclusive activity that gives more and more people with disabilities the opportunity to express themselves, exercise and improve their quality of life. In this article we present the development of a wearable sensor system capable of detecting dance gestures to command Voting, an electric wheelchair developed by the authors for dance purposes. Thus, with the support of the professional wheelchair dance teacher Gladys Foggea and the choreographer Maflohé Passedouet, thirteen dance gestures were defined, consisting of 7 simple gestures and 6 complex gestures. These gestures were used to train the algorithm of the proposed system. In order to find the appropriate algorithm and parameters for the present application, three classifiers were evaluated for their accuracy: SVM, KNN and Random Forest. Then, the most suitable parameterisation was determined by iterating each parameter for each classifier. As a result of this evaluation, it was found that the most suitable classifier was Random Forest, which achieved an accuracy of 97.7%• In addition, no difference in accuracy was observed between the detection of simple and complex gestures. Finally, the authors consider the result to be suitable to control Volting dance wheelchair, the implementation of which will be carried out in the next stage of the research.