Evaluating the accuracy of motion tracking algorithms for determining the position of finger tips

Abstract: Motion capture, or motion tracking, describes the process of recording movement of objects. Motion capture for biomechanical applications often involves sensors or markers that are placed on the skin of body segments. The position of these markers is measured as the subject moves. In this thesis, motion capture is used to record movements of the hand and fingers. Two methods for estimation of sensor positions, Method A and Method B, are investigated. These methods are based on kinematic models describing the studied system, where states are estimated with an Extended Kalman Filter. In many applications it is important to estimate the position of the finger tips accurately, but it is not always possible to place sensors on them. Therefore the methods are first modified to also estimate the positions of the finger tips based on the estimated positions of the markers. Then, the accuracy of the methods are evaluated in terms of a specific accuracy measure, for different values of the user parameters of the methods. The effect of the user parameters is investigated and appropriate values for them, i.e values that give the best performance, are given. It is shown that Method B performs better  than Method A and that it also is less sensitive to the choice of user parameters, but Method A could be considered easier to use and easier to apply to general kinematic systems.