Determination of typical patterns from strongly varying signals

Abstract

Forces measured in human joints vary considerably when an activity such as walking is carried out by different subjects or when it is repeated. 'Typical' standardised force-time patterns are needed to test and improve joint implants. Mechanically most important for their endurance are the magnitudes and times of force maxima and minima. They should equal the arithmetic means from the single measurements. Similar problems exist when evaluating other strongly varying signals, as in gait analysis. The new method to calculate typical signals (TSs) enhances existing dynamic time warping (DTW) procedures. It allows us to combine any number of signals. The sequence of input signals--used for calculating the TS--has only a minor influence. The accuracy of the method was tested numerically on signals for which the typical patterns could be defined exactly, and also on real joint forces that varied to different extents.