Due to outlier disturbance in kernel window,the traditional Mean-Shift method fails to track enlarging objects.Based on assumption that initial object model can decrease outlier disturbance on tracking results effectively,we put forward new object candidate model and similarity measure which make use of initial target model.In order to make the center and size of kernel window similar to object,a new method which can adjust dynamically to kernel window‘s size and location according to the distribution of pixel in sub-band around kernel window is also presented.The proposed method is applied to track enlarging cars,which verify the effectivity of the method.Experimental results show that the new method decreases computing complexity greatly.