Tracking of moving object using energy of biorthogonal wavelet transform

Abstract

Moving object tracking in video sequence is one of the challenging problems in computer vision. Tracking of moving object in a video is difficult due to random motion of object. Several orthogonal transform based object tracking algorithms have been proposed in literature but those methods are not able to handle object movement properly in consecutive frames due to use of non-symmetric filters. In this paper, we have proposed a moving object tracking algorithm based on the energy of the biorthogonal wavelet coefficients and the movement of object is estimated on the basis of velocity of object in consecutive frames. The approximate shift-invariance and symmetry properties of biorthogonal wavelet transform (BWT) make it useful for tracking of object in the wavelet domain. The shift-invariance property is useful for translated object modeling whereas symmetry property yields in perfect reconstruction retaining object boundaries. In addition the lifting-scheme based design of biorthogonal filters reduces computational cost. The proposed method matches computed energy of biorthogonal wavelet coefficients of object in the first frame to next consecutive frames. The computation of centroid in nth frame is predicted on the basis of centroids of three previous frames. Experimental results demonstrate the better performance of the proposed method against other state-of-the-art methods.