Kalman-mean shift tracking algorithm based on wavelet moment

Abstract

Robust real-time tracking of non-rigid objects is a challenging task. The difficulty in visual tracking is how to match targets from frame to frame quickly and reliably. Mean shift algorithm (MSA) is a typical nonparametric evaluation algorithm that needs great computation. Some scholars join Kalman filter to perform state prediction in the mean shift algorithm for reducing the computing of template matching. However, traditional Kalman filter sometimes can't track human movement very accurately because of the particularity of human joint. While wavelet moment has the multiresolution properties in addition to the invariant to the translation, scaling and rotation, so it is suitable for differentiating the details of the motion objects. Therefore, Kalman-mean shift tracking algorithm based on wavelet moment (W-K-MSA) is proposed in this paper. In this algorithm, a Kalman filter algorithm, which is used to estimate the motion parameters of targets, is improved based on wavelet moment features in the searching process. And searching window is adaptively changed, as a result, searching scope is reduced greatly, and the processing velocity and veracity is improved during model matching. The experimental results demonstrate that the proposed tracking algorithm is robust and practical.