Optimising filtering parameters for a 3D motion analysis system

In the analysis of movement data it is common practice to use a low-pass filter in order to reduce measurement noise. However, the choice of a cut-off frequency is typically rather arbitrary. The aim of the present study was to evaluate a new method to find the optimal cut-off frequency for filtering kinematic data. In particular, we propose to use rigid marker clusters to determine the dynamic precision of a given 3D motion analysis system, and to use this precision as criterion to find the optimal cut-off frequency for filtering the data. We tested this method using a model-based approach in a situation in which measurement noise is a serious concern, namely the registration of the kinematics of swimming using a video-based motion analysis system. For the model data we found that filtering the data with a single cutoff frequency of 6 Hz under some conditions decreased the accuracy of the reconstruction of the kinematics compared to using the unfiltered data. If the cut-off frequency was used that yielded optimal dynamic precision, then the accuracy improved by 29% compared to using raw data irrespective of the cluster position, close to the optimal accuracy improvement of 30%. We confirmed in an experiment that the cut-off frequency at which optimal precision was found varied between cluster positions and subjects, similar to the results obtained with the model. We conclude that 3D motion analysis systems can be made more accurate by optimising the cut-off frequency used in filtering the data with regard to their precision. Furthermore, the dynamic precision method seems useful to evaluate the effect of various filtering procedures.