Recognition of Physiotherapeutic Exercises Through DTW and Low-Cost Vision-Based Motion Capture

Abstract

Telemedicine is a current trend in healthcare. The present study is part of the ePHoRt project, which is a web-based platform for the rehabilitation of patients after hip replacement surgery. To be economically suitable the system is intended to be based on low-cost technologies, especially in terms of motion capture. This is the reason why the Kinect-based motion tracking is chosen. The paper focuses on the automatic assessment of the correctness of the exercises performed by the user. A Dynamic Time Warping (DTW) approach is used to discriminate between correct and incorrect movements. The classification of the movements through a Naïve Bayes classifier shows a very high percentage of accuracy (98.2%). Models are built for each individual and reeducation exercise with only few attributes and the same accuracy. Due to these promising results, the next step will consist of testing the algorithms on patients performing the exercises in real time.