A self-localization method entitled enhanced particle filter incorporating tournament selection and Nelder-Mead simplex search (NM-EPF) for autonomous mobile robots is proposed in this paper. To evaluate the performance of the localization scheme, an omnidirectional vision device is mounted on top of the robot to analyze the environment of a soccer robot game field. Through detecting the white boundary lines relative to the robot in the game field, weighting for each particle representing the robot's pose can be updated via the proposed NM-EPF algorithm. Because of the efficiency of the NM-EPF, particles converge to the correct location of the robot in a responsive way while tackling uncertainties. Simulation results have shown that efficiency in robot self-localization can be significantly improved while maintaining a relatively smaller mean error in comparison to that via conventional particle filter.