An Object Tracking Algorithm Based on Motion-Tuned Continuous Wavelet Transform

Abstract

A continuous wavelet transform (CWT) based object tracking algoithm is proposed. Spatio-temporal motion-tuned wavelet is used to extract motion parameters like velocity, orientation, position and scale. CWT is used to define three energy densities which are used as estimators. Sequential optimization of parameters are done in a frame-by-frame manner which allows the algorithm to track moving objects. The problem of setting initial scale parameter is improved by a new functional relation between radius of the target and velocity using a third degree polynomial constructed from 2D-Chebyshev polynomials. Experimental results show that the new functional relation gives reasonable initial scale parameter without any analysis of huge amount of previous data and the revised algorithm tracks the object in a better way.