Abstract: Simultaneous localization and mapping (SLAM) is a key technology for robots to perceive unknown environments and is a vital component in mobile robots, autonomous driving, and augmented reality. Therefore, we systematically review visual SLAM. Firstly, we briefly introduce the visual SLAM framework and analyze the difficulties and root causes of its practical implementation by analyzing trends in the volume of visual SLAM research. Secondly, we introduce the datasets and map types commonly used in visual SLAM, and analyze their specific application scenarios. Thirdly, from traditional visual SLAM to advanced visual SLAM, we introduce various algorithms for visual SLAM, as well as the improvement directions of different types of algorithms and their applicable scenarios and limitations. Further, based on traditional and advanced visual SLAM analysis, we analyze problems in visual SLAM and existing solutions. We classify current problems as adaptability issues in dynamic environments, problems caused by illumination changes, issues of insufficient closed-loop detection precision, and challenges related to computational complexity and real-time performance. At the end of each problem category, we summarize the existing solutions for each respective problem. Finally, the future trend of visual SLAM methods is discussed.