Kernel based robust object tracking using model updates and Gaussian pyramids

Visionbasedtracking, being a challenging engineering problem isone ofthehotresearch areas in*nachine vision. Inrecentstuidies Kernel based tracking usinig Bhattacharya similaritv mea.sutre is showntobean eficient techniquie fornon-rigid object tracking through thesequienceofimnages. In this paper we presented a robutst and efficient tracking approach Jbrtargets having larger motions ascompared totheir sizes. Outr tracking approach is based on calculating theGaussian pyramids ofthe imagesandthenapplying mean shift algorithm at eachpyramid level fo6r tracking thetarget. Model based tracking often sq/fkm-s abruipt changes intarget model, which iscompensated bvthemodel updatesof target. Thisleads to a very efficient androbust nonparametric tracking algorithm Thenew method is easily abletotrack thefast moving targetsandis more robulst and environment independent as compared tooriginal kernel based object tracking.