Aiming at the closedloop inaccuracy caused by perceptual ambiguity in the closedloop detection process of visual SLAM, a weighted algorithm for calculating the similarity scores of two images is proposed based on TF-IDF method, which is used for closedloop detection of visual SLAM. First, in the feature point detection, in order to obtain feature points that are evenly distributed and have fewer overlapping points, the FAST corner point detection method is used to obtain the key points, and then the key points are subjected to non-maximum suppression. Secondly, an improved algorithm is used to calculate the weighted similarity score between the two images. Finally, according to the characteristics of the scene image, closed loop confirmation is performed to further eliminate the error closed loop. The experiment tests the platform and uses the standard data set to prove that the improved closed-loop detection method can effectively improve the recognition rate and accuracy of the closed loop.