Tweet
Dear biomch-l subscribers,
We are currently comparing different methods for automated data smoothing
(selection of cut-off frequency automatically) and we would like to
receive any comments or information concerning a particular method
described in the following paper:
Simons, W. and Yang, K.H. (1991) Differentiation of human motion data
using combined spline and least squares concepts. Journal of Biomechanical
Engineering, 113, 348-351.
This main process of this method "...uses traditional least squares to fit
a cubic polynomial to SEVERAL points from the central portion of the data".
However there is no recommendation or mechanism recommended for the number
of points and choosing different number of points will lead to different
degrees of smoothing. We have implemented this algorithm in a microcomputer
and the results are similar to the test results in the paper using the data
by Pezzack et al. (1977) or the modified data by Lanshammar (1982). However,
applying this method to other data collected in the laboratory or the data
by Vaughan (1982) for example, leads to unacceptable results unless the cubic
polynomial is fitted to all the data points and not "several", corresponding
to a very low cut-off frequency. Adjusting the number of points effectively
determines the cut-off frequency unless some other method of selecting the
initial number of points is used. This however is not suggested in the
paper, indicating that the number of points to which the initial cubic is
fitted is irrelevant.
If anyone has experience in using this method as it was suggested in the
paper or modified in any way, I would appreciate it if they could send me
their comments.
Thank you very much
Dr Vasilios Baltzopoulos
Dept of Movement Science,
Faculty of Medicine,
University of Liverpool
Email: BALTZ@LIVERPOOL.AC.UK
FAX: +44 51 794 3229
We are currently comparing different methods for automated data smoothing
(selection of cut-off frequency automatically) and we would like to
receive any comments or information concerning a particular method
described in the following paper:
Simons, W. and Yang, K.H. (1991) Differentiation of human motion data
using combined spline and least squares concepts. Journal of Biomechanical
Engineering, 113, 348-351.
This main process of this method "...uses traditional least squares to fit
a cubic polynomial to SEVERAL points from the central portion of the data".
However there is no recommendation or mechanism recommended for the number
of points and choosing different number of points will lead to different
degrees of smoothing. We have implemented this algorithm in a microcomputer
and the results are similar to the test results in the paper using the data
by Pezzack et al. (1977) or the modified data by Lanshammar (1982). However,
applying this method to other data collected in the laboratory or the data
by Vaughan (1982) for example, leads to unacceptable results unless the cubic
polynomial is fitted to all the data points and not "several", corresponding
to a very low cut-off frequency. Adjusting the number of points effectively
determines the cut-off frequency unless some other method of selecting the
initial number of points is used. This however is not suggested in the
paper, indicating that the number of points to which the initial cubic is
fitted is irrelevant.
If anyone has experience in using this method as it was suggested in the
paper or modified in any way, I would appreciate it if they could send me
their comments.
Thank you very much
Dr Vasilios Baltzopoulos
Dept of Movement Science,
Faculty of Medicine,
University of Liverpool
Email: BALTZ@LIVERPOOL.AC.UK
FAX: +44 51 794 3229