An Improved Mean Shift Performance using Switching Kernels for Indonesia Vehicle License Plate Tracking Video

Abstract

In this paper, it is proposed a Mean Shift tracking uses varied kernel functions by means of switching four kernels, such as Uniform, Triangular, Epanechnikov and Gaussian for Indonesia vehicle license plate tracking. The purpose of switching kernels is to keep or to maximize the mean of the similarity function outputs which implies a successful tracking process of the vehicle license plate. The experimental results show that the average accuracy of the proposed method provides better tracking performance in term of the Average of Percentage Accuracy of Object Tracking compared to the Standard Mean Shift. Mean Shift tracking using Switching Kernel results in better tracking accuracy than standard Mean Shift with an average tracking accuracy of 71.57%.