Target Tracking Based on Wavelet Features in the Dynamic Image Sequence

Abstract

The traditional algorithm often cannot detect and extract the moving object precisely and fast, which affects result of the tracking. In order to overcome the problem, this paper proposes a method based on wavelet features for target tracking. According to the previous information obtained by Kalman filter, the possible location of the target in the frame is predicted. Multi-scale two-dimensional discrete wavelet is used to decompose the possible area in which the target exists. And then the mean and variance of the decomposed image are computed. Finally, Principal Component Analysis (PCA) is used to build an effective subspace. The target is matched to realize the tracking by measuring the similarity function. The experimental results have shown that the algorithm is robust and can improve the speed and accuracy of the target detection and tracking significantly.