With the widespread application of robot technology in complex environments, how to improve the localization accuracy and tracking stability of robots in dynamic and occluded environments has become a focal point of discussion among scholars. Traditional localization algorithms perform poorly when dealing with these complex situations. To this end, this study designs the APF-SLAM algorithm(Adaptive Particle Filter Graph Optimization SLAM) based on improved particle filtering. By introducing adaptive particle filtering and a dynamic resampling mechanism, it addresses the limitations of traditional methods in complex environments. The results show that the performance of APF-SLAM is superior to that of the other three algorithms under different environments. In static environments, the Root Mean Square Error (RMSE) is 0.38 m, and the tracking continuity is 98.4%. In dynamic environments, the RMSE of APF-SLAM is 0.42 m, demonstrating strong adaptability. Even in occluded environments, APF-SLAM can still maintain a low error propagation rate (0.08 m/s), with a localization time of only 1.8 s. In contrast, other algorithms such as EKF and G2O-SLAM exhibit large errors in dynamic and occluded environments, with poor localization accuracy and stability. The APF-SLAM algorithm excels in localization accuracy and tracking continuity in dynamic and occluded environments, effectively enhancing the robustness and real-time performance of robots in complex environments.