An adaptive weight values updating mean shift tracking algorithm

Abstract

Traditional mean shift tracking algorithm set weight value of pixels according to the distance between pixel and center of model. But it is obviously unreasonable during the tracking of asymmetric or non-rigid object, such as human. In this paper, a novel adaptive weight values updating mean shift tracking algorithm is proposed, weight value of every pixel is updated according to variation of motion state calculated by a group of Kalman filters. In this paper, this method is applied in human motion tracking, the result of experiment based on supervision video show that it has advantage on reliability and robustness.