This article presents PD-SLAM, a visual SLAM (Simultaneous Localization and Mapping) system designed specifically for dynamic environments. Accurately self-positioning and mapping the environment are crucial capabilities for intelligent robots. However, most existing SLAM systems assume static environments, which may lead to reduced robustness of the system in the presence of moving objects. PD-SLAM, built upon the ORB-SLAM3 framework, addresses this challenge by integrating a semantic segmentation module and optical flow techniques to filter out features from moving regions. Additionally, PD-SLAM utilizes a geometric feature-based method to eliminate outlier features, thereby effectively removing dynamic objects from the environment. Performance evaluations conducted on the TUM RGB-D dataset have demonstrated that PD-SLAM significantly improves absolute trajectory accuracy in dynamic environments.