Abstract: Simultaneous localization and mapping (SLAM) technology can locate itself and construct the surrounding environment in an unfamiliar environment. It has become an important basic technology in fields such as robotics, autonomous driving and virtual reality. The implicit mapping methods have a certain ability to complete and predict the unobserved areas of the scene and can realize hole filling in occluded or sparsely observed areas. Recently, it has gradually become a research hotspot that integrating the implicit mapping methods into SLAM system. This paper firstly summarized the implicit mapping methods applied in visual SLAM and classified them based on the map storage carrier. Then, it classified and explained the visual SLAM combined with implicit mapping based on improvement directions such as improving the mapping rendering speed, methods for large-scale scene expansion, enhancing the mapping robustness, improving the front-end functions and supplementing loop closure detection. It also sorted out the implicit mapping SLAM systems for specific scenarios such as semantic mapping, dynamic scenes and multi-sensor fusion. Subsequently, it introduced the commonly used datasets and evaluation criteria of implicit mapping SLAM systems and compared multiple SLAM systems based on the same datasets and evaluation criteria. Finally, it summarized the improvement methods of implicit mapping visual SLAM systems to improve their own performance, analyzed the existing shortcomings such as large computational load and serious forgetting and compared with the other technology to look ahead to the future trends.