Closed-loop detection and repositioning method based on bag of word

In order to solve the problems of increasing computational complexity and real-time system degradation caused by the continuous extension of application scenarios, this paper proposes a closed-loop detection method based on the Bag of Word model when constructing the SLAM backend optimization algorithm. The feature point set of the image set is extracted by the ORB feature extraction algorithm, and a visual dictionary tree for image similarity detection is generated. And in the closed-loop detection, the similarity detection of the image is accurately performed using the weight assignment method of the visual word. After the validation of the public data set, the improved algorithm solves the problem of scale drift of the system, improves the real-time performance of the system, and completes the relocation function of the system.