To improve the accuracy of Unmanned aerial vehicles (UAVs) performing Simultaneous Localization and Mapping (SLAM) tasks in dynamic in-door scenes, this paper proposes a visual SLAM system added line features and semantic information based on point features, termed DyPLS-SLAM. Initially, the system employs semantic segmentation to detect potential dynamic objects and conducts motion consistency checks on these objects based on the epipolar constraint. Subsequently, line features are subjected to inter-frame consistency checks based on depth information. Furthermore, keyframes significantly affected by dynamic objects are filtered based on their spatial distribution. Finally, we validated the performance of DyPLS-SLAM system using dynamic sequences of the TUM dataset, demonstrating a significant improvement in accuracy compared to classical algorithms.