Evaluating Physical Rehabilitation with a Kinect Based Low Cost System

Abstract

In this thesis a low cost low power embedded system is presented which enables the patients to evaluate the performance of their physical rehabilitation. The system uses the Asus Xtion as a motion sensor which is connected to a portable Beagleboard XM (an open source single board small computer). In order to enable the skeleton tracking functionality on the BeagleBoard XM, the Beckon SDK (provided by Omek Interactive) is used. The physical rehabilitation system uses the Beckon SDK to retrieve the joint positions from each frame and compute elbow and shoulder angles. Initially the instructor records the desired exercise and then the patient performs the same exercise. To identify the start and end of the exercise, an additional gesture has been introduced. Next, to remove redundant and noisy data from sequences, curve extraction and median filter algorithms are used. Then a comparison algorithm known as DTW (Dynamic Time Warping) is used to compare the angular sequences and compute a total score which represents a quantitative evaluation of the exercise. For visual feedback a QT interface is used. Instructions are provided on the screen to tell both the instructor and patient about what to do next. Real time angular data graphs as well as frames per second (FPS) are also displayed on screen for the user's benefit. Experiments were conducted to compare the system performance between low cost embedded systems and high end desktop machines. The skeleton tracking performance is also evaluated using a VICON motion capture system.