Closed-loop detection method based on deep learning and bag-of-words model

Abstract

The invention discloses a closed-loop detection method based on deep learning and a bag-of-word model, and the method comprises the steps: enabling a plurality of feature maps which are extracted by a VGG16 network and contain semantic information to serve as a plurality of semantic descriptors to replace ORB descriptors, and transmitting the semantic descriptors to the bag-of-word model, thereby enabling the extracted features to be more suitable for closed-loop detection; constructing a word table of the bag-of-words model, and performing clustering operation on the extracted semantic feature descriptors by using K-means to obtain a clustering center so as to serve as the word table of the bag-of-words model; extracting a feature vector under the algorithm, extracting a plurality of semantic descriptors from each image by using a VGG network, approximately replacing the semantic descriptors with words in a word list, and counting the occurrence frequency of each word in the word list in the image; and finally, calculating a similarity matrix by utilizing the feature vector. Experiments on a data set show that compared with a traditional visual bag-of-word model method, the method has higher generalization, and higher accuracy can be achieved in closed-loop detection.

Description

technical field [0001] The invention discloses a closed-loop detection method based on deep learning and a bag-of-words model, which belongs to the fields of pattern recognition, artificial intelligence and computer vision. Background technique [0002] In recent years, loop closure detection has become a key issue and research hotspot in the field of mobile robot navigation. Simultaneous localization and mapping (SLAM) is one of the key foundations for robots to move autonomously, including feature extraction and matching, data registration, loop closure detection, and global optimization. Among them, the closed-loop detection can judge whether the current location has been visited by the mobile robot, which is a key link in the SLAM process. Accurate detection of closed loops can effectively reduce the cumulative error of robot pose estimation, which is conducive to building a more accurate map and ensuring the consistency of the generated map. [0003] Some existing met...

AI Technical Summary

Problems solved by technology

On this basis, however, these methods use low-level features, which are artificially designed

Images