HOW TO FIND THE OPTIMAL CUTOFF FREQUENCY FOR FILTERING KINEMATIC DATA ?
AbstractINTRODUCTION: By analyzing movement data recent investigations distinguish between time discrete and time continuous oriented approaches (e.g. Schöner 1987, Schöllhorn 1995). Whereas the consequences of filtering procedures on instant intensities and cyclic movements are investigated expansive (Cesare et.al.), the problem on short time course characteristics of complex movement signals is still demanding. The aim of this investigation is to look for the dependance of the filter frequencies and the time course characteristics. METHODS: The data basis was formed by 2D-kinematic data of several longjumpers at the last three strides before take off. The whole movement sequence was filmed with 150Hz and lasted about 0.9sec (140fr). 20 body landmarks were digitized with an maximum spatial error of +1 cm. The data sequences of each body landmark were smoothed by means of a Butterworth-Filter with cutoff-frequencies from 0 to 20 Hz. From these filtered data as well as from the raw data the first derivations were taken by means of finite difference method. The filter caused changes were defined by the Root Mean Square Errors (RMSE) from the first time derivations of the raw and filtered data. RESULTS: Displaying the gradient of RMSE versus the filter frequencies we get a high intensity at lower frequencies and a minimum at intermediate frequencies. The optimal cutoff-frequency was obtained at the first minimum of the gradientline. At several trials of the same athlete for the same joint equal cutoff-frequencies could be found. But frequencies for different joints deviated, e.g. head: 8.5Hz, hand: 12 Hz). With different athletes different frequencies for the same joints were found, e.g. knee of athlete1: 13 Hz, knee of athlete2: 9 Hz) CONCLUSION: In order to reduce the filter caused errors, the cutoff-frequencies for the digital filter should be adapted to each athlete and each joint. By chosing the wrong filter frequency the risc is high to disregard individual movement characteristics and to manipulate the data in direction of the investigators expectations. REFERENCES: Cesare,A., Riley, P.O., Krebs, D.E. (1994). Frequency Content of Whole Body Gait Kinematic Data. Transaction on rehabilitation engineering, Vol.2,No.1. Schöllhorn,W.I. (1995). Comparison of biomechanical movement patterns by means of orthogonal reference functions. In A. Barabás, G. Fábián (Eds.), Biomechanics in Sports XII (pp. 20-24). Budapest: ITC Plantin. Schöner, G., Kelso, J. A. S. (1988). Dynamic pattern theory of behavioral change. Journal of Theoretical Biology.
Authors can retain copyright, while granting the International Society of Biomechanics in Sports (ISBS) the right of first publication.