A simple method is proposed to detect the number of change points in a sequence of independent random variables with no distributional assumption. The method is based on the weighted empirical measures over a window of observations and then runs the window over the full extent of the data. We find that the class of estimators based on our method will be consistent a.s. (almost surely) to the true number of change points and the difference between the true location of change points and the estimated location will be of order O(long) a.s. Three examples are investigated by the proposed methods.