The task of localization is a recurrent subject in the domain of autonomous robotics. Even though it is a common subject, the differences among domains causes the need of different algorithms to solve the localization problem. This paper proposes a implementation of the Monte-Carlo Localization algorithm in order to solve the localization problem for the domain of the RoboCup Humanoid Soccer KidSize. It implements methods to change the quantity of particles in function of their dispersion, to solve the robot kidnapping problem by scattering the particles and to predict the observation that leads to improvement of the confidence in the position of the robot. The experiments show that the proposed methods improved the localization process in comparison with traditional methods regarding the execution time of the algorithm and the quality of the estimated position.