Robust object tracking under appearance change conditions based on Daubechies complex wavelet transform

Abstract

In this paper, we propose a new robust algorithm for object tracking. The algorithm exploits the properties of Daubechies complex wavelet transform to track region of interest (ROI). One of the problems of using color or a similar feature for tracking is that it is sensitive to the illumination or appearance changes, so algorithms based on these features fail in presence of illumination variation or change in appearance, pose or in the presence of noise. In the proposed method, the object is represented in Daubechies complex wavelet domain to minimise the effect of frame to frame variations and noise. The reference object in the initial frame is modelled by a feature vector in terms of the coefficients of Daubechies complex wavelet transform. We investigate a similarity measure in complex wavelet domain, which makes it possible to exploit spatial information in the feature space, improving the robustness of the tracking process. An online adaptation scheme is used to update the reference object template and make the proposed tracker more robust against sudden illumination change and dramatic variation in the appearance model. The proposed tracking algorithm yields better results even in noisy video with significant variations in object’s pose and illumination.