Object tracking using particle filter in the wavelet subspace

The traditional algorithms often cannot track moving objects accurately in real time. In order to overcome the problem, this paper proposes a new method based on wavelet features for target tracking. Specifically, according to the previous information obtained by particle filter, the possible location of the target in the frame is predicted. Multi-scale two-dimensional discrete wavelet is used to characterize the possible target regions. Then the means and variances of the decomposed image are computed. Finally, Principal Component Analysis (PCA) is used to build an effective subspace. Tracking is achieved by measuring the similarity function between the target and the image regions. In addition, to combat complex background and occlusion, the characterization vector is updated based on the similarity between the object model and candidate object regions. The experimental results demonstrate that the proposed algorithm is robust and can significantly improve the speed and accuracy of target tracking.