Gu YH & Tjahjadi T
Proc. 2nd IEEE UK Symp. on Applications of Time-Frequency and Time-Scale Methods (TFTS'97), August 1997, 41-44
One problem with tracking an object through video frames and estimating the associated motion parameters is that the object of interest may undergo various affine transformations and a small deformation. This paper presents a temporal-frequency approach to solving this problem. The approach involves the use of a multiscale edge curve estimation scheme using the Hermite-Binomial filters, a robust curve feature points extraction method for generating correspondences between B-spline curve segments and for knot re-assignment, and a B-spline curve matching algorithm based on the steepest descent method for tracking moving objects and estimating affine motion parameters. The performance of the scheme has been evaluated using an image sequence of hand movements and a sequence containing a screwdriver with translation, rotation and scaling. The results show that the estimated parameters are in a very close agreement with the actual motion parameters.