Visual SLAM Loop Closure Detection Algorithm Based on LightGlue

In the traditional field of closed-loop detection, the method of manually designing features dominates. However, this method is highly susceptible to environmental variations, leading to fluctuations and inaccuracies in detection results. To overcome this limitation and improve the accuracy and robustness of loop closure detection, a method based on LightGlue is proposed. First, the SuperPoint network model is used to extract image features, obtaining a series of keypoints and corresponding descriptors. These feature information are then input into LightGlue for feature matching, generating an assignment matrix that reflects the correspondence between feature points. Finally, by calculating the similarity of the matching results, the algorithm can accurately identify the loop closure relationship between images. To validate the effectiveness of this algorithm, experiments were conducted on two widely used public datasets, NewCollege and CityCentre. Compared with other advanced loop closure detection benchmark algorithms, this algorithm demonstrates higher accuracy while maintaining a certain recall rate, showing stronger robustness and stability, and meets the basic requirements of loop closure detection.