PUB - Publikationen an der Universität Bielefeld

Studies in biomechanics, neuroethology or motor physiology often rely on kinematic analyses of the investigated movement sequence. Commercially available motion capture systems typically depend on costly hardware solutions, involving high frame rates and several camera views. Here, we introduce a low-cost motion capture algorithm that uses a manipulable kinematic model of the tracked limb or animal and stochastically optimises the model posture to match the recorded camera views. 3D movements may be analysed from a single camera view. A disambiguation method based on Markov chain analysis then determines the most likely movement sequence. Additionally to the desired joint angle time courses, the system returns local accuracy measures for each joint angle, that can be used to constrain a smoothing spline method to remove noise efficiently while preserving fast joint angle fluctuations that are real.