Smoothing of noisy human motion data using digital filtering and spline curves

Abstract

An experiment was conducted to evaluate various methods for smoothing human motion data that has been subjected to noise during the filming and digitization process. Orthogonal accelerometers were attached to the subject's leg, and records of complex dynamic kicking motions were recorded on film and simultaneously sampled through an A/D converter on an IBM PC-AT. The filmed data was hand-digitized at rates of 50 and 100 frames per second. Acceleration curves were obtained for comparison with the raw accelerometer data using finite difference techniques and direct differentiation of cubic and quintic spline curves. Pre- and postdigital lowpass filters were applied to the data in various combinations. Results of least-squares curve fits between raw and processed acceleration data suggested that excellent fit could be obtained by any of the methods if smoothing parameters were adjusted properly and if the sampling rate were high enough.<>