Mobile robot V-SLAM based on improved closed-loop detection algorithm

Aiming at the problem that the mobile robots have long-running drift in Visual Simultaneous Localization and Mapping (Visual SLAM, V-SLAM), resulting in inconsistent map construction. Proposed a VSLAM system based on improved closed-loop detection algorithm. Firstly, obtain the environmental information by the depth camera, the front-end part is completed by inter-frame registration to obtain the initial pose estimation of the robot. Secondly, the key frame is extracted by introducing the dual weights based on the degree of rotation and translation, therefor the number of comparisons is significantly reduced and the real-time performance is improved. Finally, the closed-loop detection algorithm is improved, include improved pyramid TF-IDF similarity score function is introduced to reduce the closed-loop perception ambiguity, thereby improving the accuracy of closed-loop detection; and verifying the closed-loop from the matching inner point number and the distance threshold for eliminating the wrong loop. The validity of the algorithm in the pose trajectory is verified by the standard data set, and the real-time pose can be well estimated. The correlation between the algorithm and the IAB-MAP, FAB-MAP and RTAB-MAP is more effective in the SLAM system. When the accuracy rate at 100%, the recall rate reached 76%. And the map of the environment is built in the real environment, which verifies the applicability of the actual environment of the system.