Abstract: Aiming at the problem that the map is not consistent which caused by accumulation of errors in visual simultaneous location and mapping(V-SLAM) , a visual SLAM system based on improved closed - loop detection algorithm is proposed. In order to reduce the cumulative error caused by the long operation of mobile robots, this paper introduces an improved closed-loop detection algorithm. By improving the similarity score function, it reduced the perceived ambiguity and finally made the recognition rate of closed-loop higher. To reduce the computational complexity, the environment image and depth information were directly obtained by Kinect, and the feature extraction and matching were carried out by using the small and robust ORB features. At the same time, the RANSAC algorithm was used to delete the mismatch to obtain more accurate match pairs, and then the camera poses were calculated by PnP. The more stable and accurate initial estimation pose is critical to the back-end processing, it can be attained by using the g2o to carry on unstructured iterative optimization for camera poses. Finally, in the back-end using the bundle adjustment (Bundle Adjustment, referred to as BA) as the core of the map optimization method to optimize the pose and road signs. The final experiment shows that the system can meet the real-time requirements, and can obtain a more accurate pose estimation.