A Closed-loop Detection Algorithm for Online Updating of Bag-Of-Words Model

In indoor scenes, VSLAM-based mobile robots face the challenges of poor closed-loop detection and low localization accuracy. Based on a monocular camera, we propose a closed-loop detection algorithm based on an improved real-time updating bag-of-words model. By extracting feature descriptors of online images and fusing them with loaded offline words, a fused bag of words related to the mobile robot application scenario is generated, which changes with the robot application scenario. In this paper, the improved bag-of-words and the original bag-of-words are combined with ORB-SLAM3 for closed-loop detection experiments, respectively. The experimental results show that the error between the predicted trajectory and the real trajectory of the ORB-SLAM3 system combined with the improved bag-of-words model is significantly reduced, and the robustness of the system is also improved, resulting in a certain improvement in the closed-loop detection capability of the small mobile robot.