Endpoint error in smoothing and differentiating raw kinematic data: An evaluation of four popular methods

摘要: 'Endpoint error' describes the erratic behavior at beginning and end of computed acceleration data which is commonly observed after smoothing differentiating raw displacement data. To evaluate endpoint error produced by four popular techniques, Lanshammar's (1982, J. Biomechanics 15, 99-105) modification Pezzack et al. (1977, Biomechanics, 10, 377-382) angular set was truncated three different locations corresponding to major peaks in criterion curve. Also, for each subset, padding conditions were applied. Each subset smoothed differentiated using Butterworth digital filter, cubic spline, quintic Fourier series obtain values. RMS residual errors calculated between accelerations regions. Although no method completely eliminated error, results demonstrated clear superiority spline over other methods producing accurate values close endpoints modified (1977) set. In fact, performed best with non-padded (cumulative = 48.0 rad s-2). Conversely, when applied data, filter wildly deviating more than 10 points from terminal point 226.6 better subsets padded linear extrapolation (average cumulative 68.8 s-2) analogous reflection 86.1