Currently, the application of SLAM systems in the field of navigation is quite active. However, the detection accuracy of classical SLAM systems when applied in dynamic environments is not high. This is because some dynamic objects in dynamic environments may occlude the environmental features that should have been extracted by the SLAM system. In order to solve this problem, my research conducts analysis on the detection process in dynamic environments and designs a solution: to identify and remove dynamic objects in dynamic environments to improve detection accuracy. In this research, an object detection algorithm YOLOv5 is first used to detect, identify, and remove dynamic objects in the environment extracted by the SLAM system. Then, the remaining static environmental features are passed into visual SLAM for conventional SLAM environment calculation and localization work. Finally, the modified integrated algorithm is validated and analyzed on the TUM dataset for feasibility. The results indicate that this approach successfully removes dynamic objects and effectively improves the robustness of visual SLAM applications in dynamic environments.