Abstract
Loop closure detection (LCD) is essential for the Simultaneous Localization and Mapping (SLAM) system of an autonomous robot. Aiming at the problem of false positive in traditional loop closure detection methods, in this paper we propose an end-to-end siamese network model. Firstly, the siamese network with two same branches is designed to learn the characteristics of each image pairs. Secondly, the multi-region feature weighting method is introduced to fuse saliency regions of convolutional feature map, which well reflect structured information of the environment. In the end, the geometric consistency verification is used on the candidate convolution feature map to determine the true loop. As a result, experiments on several public datasets have illustrated the superiority of our approach. Compared with traditional methods, under the precision of 100% accuracy, the recall rate is increased by 15%. Our model is more stable in different scenarios, which can achieve robustness loop closure detection.
Export citation and abstract BibTeX RIS
Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.