An efficient localization algorithm is the prerequisite for autonomous robot movement. The traditional adaptive Monte Carlo localization (AMCL) algorithm provides low pose accuracy owing to the complex environment limiting the laser model. Herein, an optimized AMCL algorithm of scan matching (SM) and discrete Fourier transform (DFT) is presented. A weighted average output of the traditional AMCL was used as the initial value of the SM, a matching function model of the lidar observation point and previous map was constructed, and the Gauss-Newton method was used to optimize the solution. Finally, the minor jitter at the localization was filtered through the DFT filter, improving the system’s stability and robustness. Through absolute localization experiments and repeated localization experiments in motion, it is verified that the optimization algorithm is superior to the traditional AMCL algorithm. The optimization algorithm effectively improves the system’s localization accuracy while maintaining its robustness.