Variational Bayesian Algorithm Based On a Wavelet For Object Tracking

In this paper, the wavelet-based variational Bayesian estimation theory for object tracking in noiseless video image sequence is used. Non-stationary signals such as images can be represented as a model through their wavelet coefficients. In this method, we suppose the mixture of normal matrices distribution over the wavelet coefficients and the variational Bayesian Expectation Maximization (VBEM) algorithm is implemented on the wavelet coefficients distribution. The wavelet coefficients are updated at each frame and we can apply them for tracking. Therefore, even where there is an obstacle in front of the object, or if noise exposure, or occlusion are found at the video, the algorithm is still able to track the target so that errors of detection and tracking in such cases are reduced.

Keywords:

object tracking , Wavelet transform , Mixture of normal matrices distribution , Variational Bayesian Inference